a

r
X
i
v
:
0
7
1
1
.
2
4
4
5
v
1


[
h
e
p
-
t
h
]


1
5

N
o
v

2
0
0
7
An introduction to quantum gravity
Bryce S. DeWitt

and Giampiero Esposito

1
1
Istituto Nazionale di Fisica Nucleare, Sezione di Napoli,
Complesso Universitario di Monte S. Angelo,
Via Cintia, Edicio 6, 80126 Napoli, Italy
(Dated: February 2, 2008)
Abstract
After an overview of the physical motivations for studying quantum gravity, we reprint THE
FORMAL STRUCTURE OF QUANTUM GRAVITY, i.e. the 1978 Carg`ese Lectures by
Professor B.S. DeWitt, with kind permission of Springer. The reader is therefore introduced, in
a pedagogical way, to the functional integral quantization of gravitation and YangMills theory.
It is hoped that such a paper will remain useful for all lecturers or Ph.D. students who face the
task of introducing (resp. learning) some basic concepts in quantum gravity in a relatively short
time. In the second part, we outline selected topics such as the braneworld picture with the same
covariant formalism of the rst part, and spectral asymptotics of Euclidean quantum gravity with
dieomorphism-invariant boundary conditions. The latter might have implications for singularity
avoidance in quantum cosmology.
1
I. MOTIVATIONS FOR AND APPROACHES TO QUANTUM GRAVITY
The aim of theoretical physics is to provide a clear conceptual framework for the wide
variety of natural phenomena, so that not only are we able to make accurate predictions to
be checked against observations, but the underlying mathematical structures of the world
we live in can also become suciently well understood by the scientic community. What
are therefore the key elements of a mathematical description of the physical world? Can we
derive all basic equations of theoretical physics from a set of symmetry principles? What do
they tell us about the origin and evolution of the universe? Why is gravitation so peculiar
with respect to all other fundamental interactions?
The above questions have received careful consideration over the last decades, and have
led, in particular, to several approaches to a theory aimed at achieving a synthesis of quantum
physics on the one hand, and general relativity on the other hand. This remains, possibly,
the most important task of theoretical physics. The need for a quantum theory of gravity
is already suggested from singularity theorems in classical cosmology. Such theorems prove
that the Einstein theory of general relativity leads to the occurrence of spacetime singularities
in a generic way [1].
At rst sight one might be tempted to conclude that a breakdown of all physical laws
occurred in the past, or that general relativity is severely incomplete, being unable to predict
what came out of a singularity. It has been therefore pointed out that all these pathological
features result from the attempt of using the Einstein theory well beyond its limit of valid-
ity, i.e. at energy scales where the fundamental theory is denitely more involved. General
relativity might be therefore viewed as a low-energy limit of a richer theory, which achieves
the synthesis of both the basic principles of modern physics and the fundamental inter-
actions in the form currently known. So far, no less than 16 major approaches to quantum
gravity have been proposed in the literature. In alphabetical order (to avoid being aected
by our own preference) they are as follows.
1. Ane quantum gravity [2]
2. Asymptotic quantization [3, 4].
3. Canonical quantum gravity [5, 6, 7, 8, 9]
2
4. Condensed-matter view: the universe in a helium droplet [10].
5. Manifestly covariant quantization [11, 12, 13, 14, 15, 16, 17].
6. Euclidean quantum gravity [18, 19].
7. Lattice formulation [20, 21].
8. Loop space representation [22, 23].
9. Non-commutative geometry [24].
10. Quantum topology [25], motivated by Wheelers quantum geometrodynamics [26].
11. Renormalization group and Weinbergs asymptotic safety [27, 28].
12. R-squared gravity [29].
13. String and brane theory [30, 31, 32].
14. Supergravity [33, 34].
15. Triangulations [35, 36, 37] and null-strut calculus [38].
16. Twistor theory [39, 40].
After such a broad list of ideas, we hereafter focus on what can be taught in a Ph.D.
course aimed at students with a eld-theoretic background. We have therefore chosen to
reprint, from section 2 to section 17, the 1978 DeWitt Lectures at Carg`ese, with kind
permission of Springer and of Professor C. DeWittMorette. In the second part we select
two related topics, i.e. the braneworld picture and the spectral asymptotics of Euclidean
quantum gravity, since they have possibly a deep impact on the current attempts to describe
a quantum origin of the physical universe. Relevant group-theoretical material is summarized
in the Appendix.
II. INTRODUCTION AND NOTATION
In 1956 Utiyama pointed out that the gravitational eld can be regarded as a non-Abelian
gauge eld. In 1963 Feynman found that in order to construct a quantum perturbation
theory for a non-Abelian gauge eld he had to introduce new graphical rules not previously
3
encountered in quantum eld theory. He showed, in one-loop order, that to preserve unitarity
one must add to every standard closed-loop graph another, involving a closed integral-spin
fermion loop. In 1966 an explicitly gauge invariant functional-integral algorithm was found
which extended Feynmans new rules to all orders (DeWitt (1967b)). A short time later it
was shown that the algorithm could be obtained by a method of factoring out the gauge
group (Faddeev and Popov (1967)).
Formally the gravitational eld and the YangMills eld can be treated identically. In
the computation of amplitudes for specic physical processes, however, the two dier by the
fact that the YangMills eld yields a renormalizable theory while the gravitational eld
does not.
Some of the proposals that have been made for dealing with quantum gravity despite its
nonrenormalizability will be discussed briey later. But it must be admitted at the outset
that we are dealing with an incomplete theory. The student may take comfort in the fact
that every formal statement will be true for all eld theories, even those, like supergravity,
possessing supergauge groups, provided they are formulated in such a way that the action
of the (super)gauge group on the eld variables is expressible without use of eld equations,
and the group operations thus given are closed.
To emphasize the generality of the formalism we shall, most of the time, suppress eld
symbols such as g

for the gravitational eld or A

for the Yang-Mills eld. The index i (or

j, k, l, etc.) will be understood to label not only a eld component but also a spacetime point
x. Thus, in the gravitational case, i will be understood to stand for the set , , x and,
in the Yang-Mills case, for the set , , x. In a supergauge theory the set i may include
spinor indices. When it does, i (or
i
) is said to be fermionic; otherwise it is bosonic.
The reason for including the continuous label x in the set i is that much of the formalism
of quantum eld theory is purely combinatorial, with summation over dummy indices being
accompanied by integration over spacetime. In order to avoid having to write a lot of integral
signs we lump x and the eld indices together and adopt the convention that the repetition
of a lower case Latin index implies a combined summation-integration. Correspondingly, a
comma followed by a lower case Latin index will denote functional dierentiation:
A
,i
A

i
. (1)
The change in a functional A (of the elds
i
) resulting from an innitesimal variation
i
4
is then
A = A
,i

i
. (2)
If A, or any of the
i
, is fermionic one must distinguish left from right dierentiation:
i,
A

i
A,
A =
i
i,
A.
Evidently
i,
A = (1)
i(A+1)
A
,i
(3)
where we adopt the rule that when an index (such as i) or a dynamical quantity (such
as A) appears as an exponent of 1 it is to be understood as assuming the value 0 or 1
according as it is bosonic or fermionic. The summation-integration convention is not to be
understood as applying to indices appearing as exponents. Such indices may participate in
summation-integrations induced by their appearance twice elsewhere in an expression, but
they temselves may not induce summation-integrations. Note that we are here treating all
quantities (A,
i
,
i
etc.) as supernumbers, i.e. as even or odd elements of an innite-
dimensional Grassmann algebra. Bosonic quantities commute with everything; fermionic
quantities anticommute among themselves. In the quantum theory this perfect commuta-
tivity or anticommutativity is broken. The corresponding quantities will then be written in
boldface.
The following notation will sometimes be convenient for expressing repeated functional
dierentiation:
...ij,
A
,kl...

j
A

l
... (4)
Note the particular examples:

i
,j
=
i
j
, (5)
(
i

j
)
,kl
= (1)
ij

i
k

j
l
+
i
l

j
k
. (6)
If x belongs to the set i and x

belongs to the set j, the generalized Kronecker delta

i
j
includes, as a factor, the spacetime delta function (x, x

).
When we are displaying specic details of a given eld theory lower case indices from
the middle of the Greek alphabet will be used to label tensor components. Coordinates in
a given local patch, or chart, will be denoted by x

, with running from 0 to n 1, n

5
being the dimensionality of spacetime. (With an eye to the ultimate application of methods
such as dimensional regularization and the renormalization group, we do not here hold n
xed at 4.) Commas followed by lower-case mid-alphabet Greek indices will denote ordinary
dierentiation with respect to the coordinates.
The following abbreviations will be useful:
g

1
2
_

_
(x, x

), (7)
A

(x, x

). (8)
It is straightforward to verify that

;
=

;
, (9)

;
=

;
, (10)
where

(x, x

),

(x, x

),
the semicolons denote covariant dierentiation, and tensor indices may be lowered and raised
by the metric tensor g

and its inverse g

respectively. (Note that we always leave the

semicolons in the lower position regardless of what happens to the indices). The delta
functions appearing above are 2-point tensor densities, or bitensor densities, of total
weight unity. In eqs. (9) and (10) the apportionment of the weight between the points x
and x

is arbitrary; in eqs. (7) and (8) all the weight is at x

. In eqs. (9) and (10) the

derivatives on the left are at x, on the right at x

.
In eqs. (8) and (10) lower case Greek indices from the beginning of the alphabet appear.
These are associated with the Yang-Mills group. The laws of covariant dierentiation of
tensors bearing various kinds of indices are determined as follows: let the symbol T represent
a tensor eld, with indices suppressed, in which we imagine all the components strung out
in a single column. Let T also be coupled to the Yang-Mills eld. Then
T
;
T
,
+ G

T + G

T, (11)
where

1
2
g

(g
,
+ g
,
g
,
), (12)
6
and the G

and G

are respectively the matrix generators of the representations of the

linear group and Yang-Mills Lie group to which T corresponds. These generators satisfy
_
G

, G

_
=

, (13)
[G

, G

] = G

, (14)
where f

are the structure constants of the Yang-Mills Lie group. When the suppressed
indices on T are restored their positions generally determine the particular representations
involved. Thus a Yang-Mills index in the upper position indicates the adjoint representation
of the Lie group and one in the lower position the contragredient representation, etc. For
physical reasons (positive probability) the Yang-Mills Lie group is required to be compact.
Therefore given representations and their contragredient forms are equivalent, and Yang-
Mills indices may be lowered (and raised) by the matrix

(and its inverse

) that
connects the adjoint and co-adjoint representations. When the Lie group is simple

may
be taken to be the Kronecker delta, and all Yang-Mills indices may be dropped to the lower
position.
We make no attempt here to list the rather complicated additional structures that appear
in supergravity theories. (The student should consult the published literature for those
details). We remark only that when spinors are present the local frame group and its spin
representations must be introduced. The local frame group is completely analogous to the
Yang-Mills group and makes a corresponding contribution to the covariant derivative on the
right side of eq. (11), with A

replaced by the connection components in the local frame

and the G

replaced by the generators of the relevant spin representation of the Lorentz

group. (See De Witt (1965) for details).
In the next section we shall show how to place the Yang-Mills group and the dieomor-
phism group (with which tensor indices are associated) on a common footing. Despite the
analogies between the two groups, as displayed for example by the similarity of the second
and third terms on the right of eq. (11), the dieomorphism group is much more complicated
than the Yang-Mills group and much less is known about its structure. Moreover, there is
a lack of symmetry between the two groups in the fact that when they are combined into
a single group (as is necessary when both Yang-Mills and gravitational elds are present)
they are united not in a direct product but in a semi-direct product based on the automor-
phisms of the Yang-Mills group under dieomorphisms. The same is true for the combined
7
dieomorphism and local frame groups, when spinor elds are present.
Our list of notational conventions is completed with the following statements and deni-
tions:
T
;
T
;
=
_
G

+ G

_
T, (15)
R

,
+

, (16)
F

,
A

,
+ f

, (17)
R

, R R

. (18)
Unless otherwise specied we shall assume that spacetime is globally hyperbolic and com-
plete
1
. Without loss of generality x
0
may then be assumed to be a global time coordinate,
in the sense that it denes a foliation of spacetime into smooth complete hypersurfaces
x
0
= constant, arranged in a temporal order. These hypersurfaces need not be everywhere
spacelike, although if they are noncompact they must be asymptotically spacelike. The
signature of the metric tensor will be + ++ ..., and units (when needed) will be chosen
to be absolute, with h = c = 32G = 1. The absolute units of length, time and mass
respectively are 1.6 10
32
cm., 5 10
43
sec. and 2 10
6
g., which gives an idea of the
domains in which quantum gravity becomes relevant.
III. THE GAUGE GROUP
The gauge group of quantum gravity is the dieomorphism group and that of Yang-Mills
theory is the Yang-Mills group. We begin with the latter.
Elements of the Yang-Mills group are locally parametrized by a set of dierentiable scalar
functions

(x), with

= 0 denoting the identity element. Elements innitesimally close

to the identity are parameterized by innitesimal scalars. The action of such an element on
the Yang-Mills potentials is given by
A

,
+ f

;
, (19)
the covariant derivative being determined by noting that

transforms (under inner auto-

morphisms) according to the adjoint representation of the group. It will be convenient to
1
There is some evidence (although hardly overwhelming yet) that quantization suppresses the singularities
in spacetime that often develop automatically in classical general relativity.
8
rewrite eq. (19) in the form
A

=
_
Q

d
n
x

, d
n
x

n1

=0
dx

, (20)
or, in the generic notation,

i
= Q
i

, (21)
where
Q

;
. (22)
In passing from eq. (20) to the generic form (21) one replaces the labels , , x by the index
i and the labels

, x

by the index , and one understands that repetition of the latter index
implies a combined summation-integration.
In quantum gravity the action of the dieomorphism group can be expressed in identical
generic form. The dieomorphism group is the group of mappings f : M M of the space-
time manifold M into itself such that f is one-to-one and both f and f
1
are dierentiable.
In practice one may require f and f
1
to be C

and, if the sections x

0
= constant are non-
compact, to reduce asymptotically (i.e. at spatial innity) to the local identity mapping.
Such mappings dene a dragging of all tensor elds dened on M, and if the mapping
is innitesimally close to the identity the dragging may be viewed as a physical displace-
ment of the elds through an innitesimal vector . If all elds, including the metric (i.e.
gravitational) eld, are displaced by the same amount the physics remains unchanged. It
is conventional in physics, therefore, to adopt an opposite viewpoint and to regard an in-
nitesimal dieomorphism as leaving the physical points of the manifold untouched while
dragging all coordinate patches (i.e. the complete atlas) through the negative vector .
Locally this is expressed by the coordinate transformation x

where

= x

. (23)
Let T be a tensor eld and T its change under dragging through . The Lie derivative
of T with respect to is dened by
L

T = T. (24)
Let p be a point of M and p

the point to which it is dragged under . Then the coordinates

of p in the new coordinate system (23) are identical with those of p

in the old coordinate

9
system. Moreover, the components of T at p in the new coordinate system are identical
with those of T + T at p

in the old coordinate system. One has only to cast eq. (24) into
component language, therefore, to regard the Lie derivative as expressing the negative of
the change in the functional form of the components of T, viewed as functions of the
local coordinates, under the dieomorphism. This enables one to compute
L

T = T
,

,
= T
;

;
, (25)
which yields, in particular, the gauge transformation law for the metric tensor:
g

=
_
L

g
_

=
;

;
. (26)
Equation (26) may be rewritten
g

=
_
Q

d
n
x

, (27)
Q

;
, (28)
which, if the labels , , x are replaced by i and the labels

, x

by , takes the generic form

(21).
Lower case Greek indices from the rst part of the alphabet, as in eq. (21), will from
now on be called group indices. If the group indices are allowed to label fermionic as
well as bosonic gauge parameters then the generic form (21) holds also for the supergauge
transformations of supergravity theories. Although we shall not go into the specic details
of such theories we shall, in all that follows, allow for their possible presence.
IV. STRUCTURE CONSTANTS
By invoking the requirement that the commutator of two innitesimal gauge group op-
erations be itself a group operation (the closure property) one arrives at the functional
dierential identity
Q
i
,j
Q
j

(1)

Q
i
,j
Q
j

= Q
i

, (29)
where the Cs are certain coecients known as the structure constants of the gauge group.
They possess the symmetry
C

= (1)

. (30)
10
The structure constants of the Yang-Mills group may be determined by straightforward
computation from Eq. (19), (20) and (22). They are the components of the following 3-point
tensor density:
C

= f

(x, x

)(x, x

). (31)
The weights are at x

and x

.
The structure constants of the dieomorphism group may be determined by recalling the
commutation law for the Lie derivative:
_
L
X
, L
Y
_
T = L
[X,Y ]
T. (32)
Here [X, Y ] is the Lie bracket of the vectors X and Y :
[X, Y ] = L
X
Y = L
Y
X. (33)
The structure constants are the components of the 3-point tensor density dened by
_
d
n
x

_
d
n
x

= [X, Y ]

= X

,
Y

Y

,
X

= X

;
Y

;
X

. (34)
Evidently
C

. (35)
The weights are at x

and x

.
The action of the gauge group on the eld variables
i
, expressed by eq. (21), is a
realization of the group. This realization is always a faithful one, which implies that
Q
i

= 0 for all i if and only if X

= 0 for all . By functionally dierentiating eq. (29)

with respect to
k
, multiplying by Q
k

, judiciously permuting the indices , , , adding

the results, and invoking the faithfulness of the realization, one obtains the following cyclic
identity satised by the structure constants:
C

+ (1)
(+)
C

+ (1)
(+)
C

= 0. (36)
In the case of the Yang-Mills group this identity reduces to the corresponding identity for
the constants f

. In the case of the dieomorphism group it is the Jacobi identity for Lie
brackets.
11
V. CONFIGURATION SPACE. ORBITS
For each point x in the spacetime manifold M, the eld (index i suppressed) takes
its value in a certain nite-dimensional dierentiable manifold
x
, which may but need
not be a vector space or subspace thereof. In pure gravity theory, for example,
x
is the
subspace of Sym
_
T

x
T

x
_
containing all local symmetric covariant second rank tensors
at x having nonvanishing determinant and signature + + + .... Here T

x
is the dual of
the tangent space to M at x, and Sym denotes the symmetric part of the tensor product
T

x
T

x
.
The set of all
x
with x in M, may be regarded as forming a ber bundle over M. Each

x
is a ber, and each eld is a cross section of the bundle
2
. The bundle may but need
not be a simple product bundle.
The set of all cross sections, i.e. of all eld congurations may be assembled into a
space called the conguration space. Because of dierentiability requirements on the
eld congurations is endowed naturally with a functional dierentiable structure and
may be viewed as an innite dimensional dierentiable manifold, or, if
i
includes fermion
elds, as a dierentiable supermanifold (also known as a Z
2
-graded manifold. See Kostant
(1977)). Since M is never compact (at least in the time direction) the elds are usually
constrained to obey also special boundary conditions at innity. In both Yang-Mills and
gravity theory these can be of considerable importance.
Let the gauge group be denoted by G and let be an element of G. Denote by

the
point of to which is displaced under the action of . The set of points

for all in G
is known as the orbit of and denoted by Orb.(). The set of all orbits can be assembled
into a space called the space of orbits, denoted by the quotient symbol /G. Since all elds
on a given orbit describe the same physics it is the space of orbits that constitutes the real
physical conguration space of the theory. In pure gravity theory is the space, Lor(M),
of Lorentzian (also called pseudo-Riemannian) metrics on M, G is the group, Di(M), of
dieomorphisms of M, and the physical conguration space, Lor(M)/Di(M), is the space
of Lorentzian geometries on M.
Because gauge groups can be coordinatized by dierentiable functions (i.e. the gauge
2
If the ber bundle admits no global cross sections the eld must be dened by introducing overlapping
patches.
12
parameters) G, like , can be regarded as an innite dimensional dierentiable manifold
(or supermanifold). In Yang-Mills theory G may have a simple product structure inherited
from the associated Lie group of the theory (see (31)), or it may itself be a twisted bundle.
The dieomorphism group of gravity theory, by contrast, cannot be viewed as a bundle but
has a structure that is much less well understood. Some, but only a little, of its complexity
will emerge as we go along.
Since both and G are dierentiable (super) manifolds the quotient space /G too is
a dierentiable (super) manifold, or rather it is a dierentiable (super) manifold that may
have a boundary.
To see how a boundary can arise consider a typical, i.e. generic, orbit. Modulo a
possible discrete center it is a copy of G, because it provides a realization of G and has the
same dimensionality. Not all orbits need have this dimensionality. There is often a class of
degenerate orbits having fewer dimensions. These are the orbits that remain invariant under
the action of nontrivial continuous subgroups of G. They are the boundary points of /G.
To see this think of as being R
3
and G as being the group of rotations about a xed axis.
The orbits are then circles perpendicular to and centered on the axis, and the orbit space
is a half-plane whose boundary points correspond to the points on the axis, which remain
invariant under the group.
The greater the dimensionality of the subgroup that leaves a given orbit invariant, the
smaller the dimensionality of the orbit. Fischer (1970) has shown that if the invariance
group has only one dimension then the orbit is an ordinary boundary point of /G. If the
invariance group has two dimensions then the orbit lies on a boundary of the boundary and
so on. The whole orbit manifold, with its boundary, and its boundaries of boundaries, etc.,
is known as a stratied manifold.
In gravity theory the boundary orbits are the symmetrical geometries, i.e. those that
possess Killing vectors. The boundary structure of Lor(M)/Di(M) in general depends
critically on M. Since there exists no complete classication of n-dimensional manifolds
(n > 3) that can possess globally hyperbolic metric tensors, there exists also no complete
classication of possible conguration spaces for the gravitational eld. On some space-
times there may be no Lorentzian geometries possessing Killing vectors. Such spacetimes
are called wild. If M is wild Lor(M)/Di(M) has no boundary points. For technical rea-
sons it is frequently necessary to regard certain familiar spacetimes as wild. For example,
13
asymptotically at spacetimes dieomorphic to R
n
are usually treated as wild. The reason
for this is to keep Lorentz transformations distinct from gauge transformations, by requiring
the gauge parameters

to vanish at innity. Flat Minkowski spacetime is then not a

boundary point of Lor(R
n
)/Di(R
n
) because the Poincare isometries are not regarded as
being contained in Di(R
n
).
VI. METRICS ON CONFIGURATION SPACE
It turns out to be both possible and useful to regard and /G not merely as dier-
entiable (super)manifolds but as pseudo-Riemannian (super)manifolds as well. Let d
i
be
an innitesimal displacement in . We may associate with this displacement a (super) arc
lengths ds, given by
ds
2
= d
i
i

j
d
j
, (37)
where
i

j
are the components of a (super) metric tensor on . The
i

j
are functionals of
having the symmetry
i

j
= (1)
i+j+ij
j

i
and forming an invertible continuous matrix. The inverse, denoted by
ij
, satises
i

kj
=
j
i
,
ik
k

j
=
i
j
,
ij
= (1)
ij

ji
. (38)
If the metric
i

j
is chosen in such a way that the actions of G on are isometries then
i

j
induces also a metric on the orbit space /G. One simply denes the distance between
neighbouring orbits in /G to be the orthogonal distance between them in . This requires
selecting
i

j
in such a way that the continuous matrix (1)
(i+1)
Q
i

j
Q
j

is nonsingular,
on all orbits so that a vector cannot be simultaneously tangent to and orthogonal to any of
them.
Every element of G innitesimally close to the identity generates a vector eld Q
i

on (see eq. (21)). Each of these elds is a linear combination of the basic vector elds
Q
i

, and the structure of G is determined by the (super) Lie bracket relations (29) that they
satisfy. The condition that G acts isometrically on may be translated into the statement
that the (super) Lie derivatives of the metric
i

j
with respect to the Q

s all vanish:
0 =
i

L
Q
=
i

j,k
Q
k

+ (1)
(j+k)
i,
Q
k

j
+ (1)
j
i

k
Q
k
,j
. (39)
14
It is not dicult to verify that eq. (29) is the integrability condition for (39). Equation
(39) generally has an innity of solutions diering nontrivially from one another. If it did
not, i.e. if the solution were unique up to a constant factor, this would mean that G acts
transitively on and hence that /G is trivial, the theory having no physical content.
In order to understand what eq. (39) says in more familiar terms it is helpful to note
that the elds
i
encountered in practice usually provide linear realizations of their gauge
groups. This is cetainly true for the Yang-Mills and gravitational elds. What it means
is that the functional derivatives Q
i
,j
are independent of the
i
and, when regarded as
continuous matrices (in i and j), yield a matrix representation of the (graded) Lie algebra
associated with the group.
Of course this simplicity is generally lost if the s are replaced by nonlinear functions
of themselves. But it is remarkable that there is usually a natural set of eld variables of
which the Qs are linear functionals. In gravity theory, in fact, there is a family of natural
elds, namely all tensor densities of the form
(

g
r
g

or (

g
r
g

, r ,= 1/n, (40)
where g det(g

).
Consider now the way in which the Qs themselves change under innitesimal gauge
transformations. Using eq. (29) one nds
Q
i

= Q
i
,j

j
= Q
i
,j
Q
j

= (1)

_
Q
i
,j
Q
j

Q
i

, (41)
which says that Q
i

is a two-point function that transforms at the point associated with i

according to the representation generated by the matrices
_
Q
i
,j
_
and at the point associated
with contragrediently to the representation generated by the matrices
_
C

_
. In gravity
theory this says that the function Q

of eq. (28) transforms like a covariant tensor at x

and like a covariant vector density of unit weight at x

,
3
which indeed it does. Equation (41)
yields an analogous statement about the transformation law of the two-point function Q

of eq. (22) under the Yang-Mills group.

3
Under the dieomorphism group a tensor density of weight w, having p covariant and q contravariant
indices, transforms contragrediently to a tensor density of weight 1 w, having q covariant and p con-
travariant indices.
15
We are now ready to interpret eq. (39). Under the gauge group the s change according
to

j
=
i

j,k

k
=
i

j,k
Q
k

= (1)
j
_
(1)
k
i,
Q
k

j
+
i

k
Q
k
,j
_

, (42)
which says that the s are two-point functions that transform at each point contragrediently
to the representation generated by the matrices
_
Q
i
,j
_
. In gravity theory this implies that
when the
i
are chosen to be the components of the covariant metric tensor,
i

j
must
transform at each point like a symmetric contravariant tensor density of unit weight. Any
s that transform in this way, and have an inverse
ij
, provide an acceptable metric on
Lor(M).
Among all such metrics on Lor(M) there is a unique (up to a constant factor) 1-parameter
family of them that may be characterized as local. These are given by

(x, x

), (43)

1
2
g
1/2
_
g

+ g

+ g

_
, ,=
2
n
. (44)
For Yang-Mills theory in at spacetime the corresponding metric is

(x, x

), (45)
where

is the Minkowski metric. The matrices (1)

(i+1)
Q
i

j
Q
j

in the two cases are

readily calculated to be respectively
2g
1/2
_

;
+ R

(1 +)[(x, x

)]
;

_
(46)
and

;
. (47)
The continuous matrix (47) is eectively the negative of the Yang-Mills-invariant Laplace-
Beltrami operator. If the Yang-Mills eld is untwisted (see the lectures by Avis and Isham
in this volume) it is a nonsingular operator having a unique Greens function for each choice
of boundary conditions at innity. If the Yang-Mills eld is twisted, however it may have
zero eigenvalues, which means that the choice (45) fails to yield a globally valid metric on
the orbit manifold. Although this is an important and interesting situation we shall not
attempt to deal with it in these lectures.
16
The continuous matrix (46) too may become singular. Its structure is simplest when
= 1(n ,= 2); it is then eectively a slightly generalized form of the standard Laplace-
Beltrami operator. Considerable evidence exists to indicate that it is nonsingular when
the spacetime manifold is dieomorphic to R
n
. But for other topologies it may have zero
eigenvalues. Again we exclude this situation from consideration.
When the matrices (46) and (47) are nonsingular, expressions (43) and (45) constitute
metrics on the space of elds which dene, by orthogonal projection, globally valid non-
singular metrics on the space of orbits. It is possible to develop a theory of geodesics on
these conguration spaces. The geometry dened by (45) is at and the geodesics in the
space of Yang-Mills potentials are trivial. The geometry dened by (43) and (44), on the
other hand, is not at, and the resulting theory of geodesics on the space of metric tensors
g

is not trivial. It can nevertheless be shown that any pair of points in this space can be
connected by a unique geodesic. It can also be shown that if a geodesic intersects one orbit
orthogonally then it intersects every orbit in its path orthogonally, and, moreover, traces out
a geodesic curve in the space of orbits. Methods for proving these theorems can be found in
DeWitt (1967a). Using these theorems together with the fact that a vector in the space of
metric tensors cannot be simultaneously parallel and orthogonal to an orbit, one can then
prove that any pair of orbits can be connected by a unique geodesic. It should be stated
that all of these theorems depend upon the maintenance of xed boundary conditions (on
elds and dieomorphisms) at innity.
VII. VOLUME ELEMENTS ON CONFIGURATION SPACE
With a metric dened on the space of elds it is possible to introduce a formal volume
element d (d

i
d
i
) by choosing
= const. [det(
i

j
)[
1/2
. (48)
This volume element is gauge invariant and can be used to dene gauge invariant functional
integrals over conguration space. When fermionic elds are present the determinant in eq.
(48) is the super determinant (see Nath (1976)), which satises the variational law
ln det(
i

j
) = (1)
i

ij

i
. (49)
17
This law, combined with eq. (39), yields the following equation of divergenceless ow that
could in principle be used to select a gauge invariant volume element independently of a
metric:
(1)
i(+1)
_
Q
i

_
,i
= 0. (50)
The delta functions contained in the metrics (43) and (45) give these metrics a block
structure that yields simple formal expressions for their determinants. In Yang-Mills theory
the determinant is a constant; in gravity theory it is given by
det
_

_
=

x
(x), (51)
where (x) is the determinant of the
1
2
n(n+1)
1
2
n(n+1) matrix

. It is not a dicult
computation to show that
= (1)
n1
_
1 +
n
2
_
g
1
4
(n4)(n+1)
. (52)
In a 4-dimensional spacetime , and hence det(

), is seen to be a constant, independent

of the g

. The functional , in the volume element over the conguration space of gravity
theory, may therefore be taken to be a constant. Without loss of generality it may be chosen
equal to 1. This will no longer be true in other dimensions, or when other elds are present in
addition to the gravitational eld, if we stick to the g

as the basic eld variables. However,

we can in principle replace the g

by one of the family of variables dened in eq. (40) and

choose r so that remains constant. In practice, as we shall see later, this is unnecessary.
To set eectively equal to unity it turns out to be necessary only to choose basic elds
that transform linearly under the gauge group.
VIII. GROUP COORDINATES
The scalar functions

(x) that parameterize the elements of the Yang-Mills group may

be regarded as coordinates in the group manifold. In the case of the dieomorphism group
the group coordinates may be taken to be the functions

(x) that dene the coordinate

transformation x

associated with each dieomorphism in each coordinate chart or

patch. Note that the functions

(x) are neither scalars nor components of vectors. Note also

that in both cases the coordinatization of the group cannot generally be achieved without
18
bringing in the whole apparatus of charts, atlases and consistency conditions in the regions
of intersection of overlapping charts.
Any group may be regarded as acting on itself through multiplication either on the left or
on the right. Every group thus provides a realization of itself, and if it is a gauge group G,
possesses a set of functionals Q

[] analogous to the functionals Q

i

[] over the conguration

space . The Q

are then dened by

[(I + )]

+ Q

[]

for all

and all in G, (53)

where I denotes the identity element of G and I + denotes an element of G whose
coordinates dier by innitesimal amounts

from those of I. Using the fact that I

(x) =
x

, and that (

(x) =

((x)) for all and

in G, it is not dicult to verify that the

Q

for the dieomorphism group are given explicitly by

Q

[] =

((x), x

). (54)
In the case of the Yang-Mills groups the Q

take the forms

Q

[] = g

((x))(x, x

), (55)
where the g

are the corresponding quantities for the associated Lie group.

Now let

t for some xed

and consider the curve in G dened by

(t) = lim
t0
(I + )
t/t
. (56)
Evidently
(s)(t) = (t)(s) = (s + t), (0) = I,
1
(t) = (t), (57)
d

(t)
dt
= Q

[(t)]

. (58)
The points on the curve are seen to constitute a one-parameter Abelian subgroup of G.
If it could be proved that all the elements of G in a neighbourhood N of the identity can
be obtained by a process of exponentiation of the form (56) then it would follow that the
one-parameter Abelian subgroups completely span the neighbourhood N. A special set of
coordinates

c
, known as canonical coordinates, could be introduced in N for which the
functions

(t) above take the simple form

c
(t) =

t. (59)
19
Let us assume that our coordinates are already canonical, so that we may drop the subscript
c. Then we have
I

= 0,
1

, (60)

= Q

[]

= Q
1

[]

, (61)
where
_
Q
1

_
is the (continuous) matrix inverse to
_
Q

_
. By taking note of the fact that
the Q

must satisfy an identity analogous to (29), namely

Q

,
Q

(1)

,
Q

= Q

, (62)
we may show that in a canonical coordinate system the Q

are completely determined by

the structure constants.
We begin by rewriting eq. (62) in the equivalent form
Q
1
,
(1)

Q
1
,
+ (1)
(+)
C

Q
1

Q
1

= 0. (63)
Multiplying this equation on the right by

and using eq. (61) we get

Q
1
,

(1)

Q
1
,

+ C

Q
1

= 0. (64)
On the other hand, dierentiating eq. (61) with respect to

we nd
(1)

Q
1
,

+ Q
1

. (65)
Addition of eqs. (64) and (65) yields
Q
1
,

+ Q
1
C Q
1
= 1, (66)
where 1 denotes the unit matrix (delta function) and
Q
1
[]
_
Q
1

[]
_
, C (1)

=
_
C

_
. (67)
The solution of eq. (66) satisfying the necessary boundary condition
Q

[I] =

(68)
(see eq. (53)) is
Q
1
[] =
e
C
1
C
1 +
1
2!
C +
1
3!
(C )
2
+ ... . (69)
20
The series (69) converges for all values of the

. For certain values the (continuous)

matrix Q
1
may have vanishing roots. For these values some of the Q

, and hence the

canonical coordinate system itself, become singular. In the case of an untwisted Yang-Mills
group it can be shown that the one-parameter Abelian subgroups do span a neighbourhood
of the identity. (This is, in fact, a corollary of the corresponding theorem for the associated
Lie group.) Indeed they span the entire group - or, rather, that part of the group that
is connected to the identity, i.e. the proper group. The whole group can therefore be
parameterized by canonical coordinates (supplemented, perhaps, with some discrete labels).
Canonical coordinates for the Yang-Mills group have a periodic, or angular, nature. At
a given point x of the spacetime manifold let the

(x) in eq. (55) increase in magnitude

but maintain xed ratios to one another. Eventually all of the g

will become singular at

once. One has returned to the identity element of the associated Lie group. By allowing
the canonical coordinates to range from to one evidently covers the gauge group an
innity of times. Despite the fact that the Q

become singular for certain values of the

s,
the canonical coordinates are good in that, no matter what their values, they always dene
a unique element of the group.
IX. NO CANONICAL COORDINATES FOR THE DIFFEOMORPHISM GROUP
If canonical coordinates could be introduced into the dieomorphism group one could
dispense with the apparatus of charts, atlases, etc. in parameterizing the group. Every
dieomorphism could be characterized by a (nite) vector eld just as those innitesimally
close to the identity can be characterized by an innitesimal vector eld. And a vector eld
has a meaning independent of charts and atlases.
Unfortunately the one-parameter Abelian subgroups of the dieomorphism group do not
span a neighbourhood of the identity. If the dimensionality n of the spacetime manifold M
is greater than or equal to 2 there are C

dieomorphisms arbitrarily close to the identity

that cannot be obtained by exponentiation as in eq. (56). The proof, which we now outline,
was rst given by Freifeld (1968).
It suces to conne attention to R
2
or, equivalently, to the complex plane C. Let x be
a point of C. Instead of breaking x into its real and imaginary parts we may treat x and its
complex conjugate x

formally as independent variables. A C

dieomorphism : C C
21
is then a one-to-one complex function (x, x

), of class C

in both x and x

, whose inverse,
x(,

) is C

in and

.
Let N be a positive integer and a positive real number. Suppose has the analytic
form
(x, x

) = e
2i
N
x + x
N+1
(70)
in a nite neighbourhood of the origin (e.g., in a circle of nite radius), and suppose that
outside of this neighbourhood changes smoothly (C

) to the identity function (x, x

) = x.
If N is chosen large and is chosen small then and all its derivatives may be made uniformly
close to those of the identity. We shall show that does not lie on a one-parameter subgroup
of C

dieomorphisms (t) : C C with (0) = I.

Suppose we assume that it does lie on such a subgroup. Without loss of generality we
may also assume that (1) = , and then we have
(0, x, x

) = x, (1, x, x

) = (x, x

), (71)
as well as
(s, (t, x, x

),

(t, x, x

)) = (t, (s, x, x

),

(s, x, x

)) = (s + t, x, x

). (72)
Note that the dieomorphism (70) leaves the origin xed. Therefore
(0, 0, 0) = 0, (1, 0, 0) = 0. (73)
Dene
z(t) (t, 0, 0). (74)
The function z(t) describes a closed curve passing through the origin in the complex plane.
Using eqs. (72) and (73) we nd
(z(t), z

(t)) = (1, z(t), z

(t)) = (1, (t, 0, 0),

(t, 0, 0))
= (t, (1, 0, 0),

(1, 0, 0)) = (t, 0, 0) = z(t), (75)

which implies that the dieomorphism (70) leaves every point on this closed curve xed.
But the only curve passing through the origin that (70) leaves xed is the degenerate curve
consisting of the single point x = 0. Therefore every one of the dieomorphisms (t) must
leave the origin xed:
(t, 0, 0) = 0 for all t. (76)
22
Since and the (t) are C

we may consider their formal Taylor series at the origin.

The formal Taylor series for , which is just expression (70), must lie on the one-parameter
group of formal Taylor series for the (t), which may be written in the form
(t, x, x

) =

m,n=0
a
m,n
(t)x
m
x
n
. (77)
Furthermore, these formal Taylor series must satisfy (formally) eqs. (72).
In view of eqs. (71) and (76) it is evident that
a
0,0
(t) = 0 for all t;
a
1,0
(0) = 1, all other a
m,n
(0)

s vanish;
a
1,0
(1) = e
2i
N
, a
N+1,0
(1) = , all other a
m,n
(1)

s vanish. (78)
Moreover, inserting (77) into (72) with s = t =
1
2
, one nds
e
2i
N
x + x
N+1
=

m,n=0
a
m,n
_
1
2
__

_
1
2
, x, x

__
m
_

_
1
2
, x, x

__
n
= a
1,0
_
1
2
__
a
1,0
_
1
2
_
x + a
0,1
_
1
2
_
x

+ ...
_
+ a
0,1
_
1
2
__
a

1,0
_
1
2
_
x

+ a

0,1
_
1
2
_
x + ...
_
+ ...,
whence
e
2i
N
=
_
a
1,0
_
1
2
__
2
+[a
0,1
(1/2)[
2
, (79)
0 = a
0,1
_
1
2
_
Rea
1,0
_
1
2
_
. (80)
Suppose a
0,1
_
1
2
_
,= 0. Then a
1,0
_
1
2
_
must be pure imaginary, and the right hand side of eq.
(79) must be a real number, which contradicts the left hand side. Therefore
a
0,1
_
1
2
_
= 0, a
1,0
_
1
2
_
= e
1
2
(
2i
N
+2iK)
,
for some integer K. Repeating this reasoning for s = t =
1
4
, s = t =
1
8
, etc., one obtains, by
continuity,
a
0,1
(t) = 0 for all t, a
1,0
(t) = e
t
, = 2i
_
1
N
+ K
_
. (81)
We now have
(t, x, x

) = e
t
x +

m+n2
a
m,n
(t)x
m
x
n
. (82)
Insertion of this formal series into (72) yields
a
m,n
(s + t) = e
s
a
m,n
(t) + e
(mn)t
a
m,n
(s), m+ n = 2. (83)
23
This functional equation can be solved by dierentiating with respect to s and setting s = 0:
_
d
dt

_
a
m,n
(t) = a
m,n
(0)e
(mn)t
, m+ n = 2. (84)
(Here the dot denotes the derivative.) The Greens function for the operator
d
dt
appro-
priate to the boundary conditions (78) is
_
(t t

)(t

) (t

t)(t

)
_
e
(tt

)
where is the step function. Use of this Greens function yields
a
m,n
(t) =
a
m,n
(0)
(mn 1)
_
e
(mn)t
e
t
_
, m+ n = 2, (85)
which is easily veried to satisfy (83). Now if N is large a
m,n
(1) (m + n = 2) must vanish
in virtue of the last of eqs. (78). But the right hand side of (85) does not vanish at t = 1
unless a
m,n
(0) = 0. Therefore
a
m,n
(t) = 0 for all t when m + n = 2, (86)
and hence
(t, x, x

) = e
t
x +

m+n3
a
m,n
(t)x
m
x
n
. (87)
Inserting this series into (72) one gets
a
m,n
(s + t) = e
s
a
m,n
(t) + e
(mn)t
a
m,n
(s), m+ n = 3, (88)
which is identical with eq. (83) except that now m + n = 3. The solution is the same as
before:
a
m,n
(t) =
a
m,n
(0)
(mn 1)
_
e
(mn)t
e
t
_
, m+ n = 3. (89)
It is now possible for the factor mn 1 in the denominator to vanish, in which case this
solution is replaced by its limit as mn 1:
a
m,n
(t) = a
m,n
(0)te
t
, mn = 1. (90)
Once again, comparing (89) and (90) with the boundary condition a
m,n
(1) = 0 (m+n = 3),
one must conclude that
a
m,n
(t) = 0 for all t when m+ n = 3. (91)
24
In fact, continuing in this way one nds
a
m,n
(t) = 0 for all t and all m, n with 2 m + n N. (92)
One arrives nally at the case m + n = N + 1, where one obtains
a
N+1,0
(t) =
1
N
a
N+1,0
(0)e
t
_
e
Nt
1
_
. (93)
This expression vanishes at t = 1, precisely where we dont want it to! According to (78)
we must have a
N+1,0
(1) = . We have thus arrived at a contradiction. Q.E.D.
Since only a small (innitesimal) neighbourhood of the origin is really involved in the
above analysis, it follows that the restriction to R
2
is not essential. For any dierentiable
manifold of dimension greater than or equal to 2 there exist C

dieomorphisms arbitrarily
close to the identity that do not lie on one-parameter subgroups of C

dieomorphisms.
X. GAUGE CONDITIONS
A gauge condition is a set of constraints that picks out a subspace in the conguration
space , of codimension equal to the dimension of the gauge group G. The gauge condition
is said to be globally valid if this subspace intersects each orbit in precisely one point. Such
a subspace exists in the Yang-Mills case only if the gauge group corresponds to an untwisted
ber bundle. For the dieomorphism group it probably exists if spacetime is dieomorphic
to R
n
. We conne our attention to these cases. The subspace may then be regarded as
representing the orbit manifold /G. Each orbit is represented by the point at which it
intersects the subspace.
To express this idea in equations one may think of the variables
i
as being replaced by
other variables I
A
, P

, where the I
A
label individual orbits and are gauge invariant, and the
P

label corresponding points in each orbit. The point on each orbit that is selected by the
given gauge condition may be chosen as the origin of the coordinates P

in that orbit.
The gauge condition is then simply P

= 0.
It will actually prove convenient to work with the continuum of gauge conditions
P

[] =

, (94)
where the

are constants (i.e., independent of the

i
) whose values range over some
preselected domain. Explicit functional forms for the P

in terms of the
i
may be obtained
25
(in principle) as follows. Remembering that each (generic) orbit is a copy of G, choose the
P

to be a set of group coordinates. Since the action of the gauge group on each (generic)
orbit mimics its action on itself such Ps must be solutions of the functional dierential
equations
4
P

,i
[]Q
i

[] = Q

[P[]]. (95)
The domain over which the

in eq. (94) range may then be taken to be the full domain of

the group coordinates.
Equations (95) do not suce completely to determine the P

. Additional conditions
are needed to line up corresponding points on adjacent orbits. One possible way to do
the lining up is as follows. Introduce into the conguration space one of the metrics
i

j
previously discussed. Choose a generic orbit and call it the base orbit. Call the identity
element on that orbit the base point. Let V be the subspace of generated by the set of
all geodesics emanating from the base point in directions orthogonal to the base orbit. As
previously noted, these geodesics intersect all orbits in their paths orthogonally. Using the
fact that every pair of points in can be connected by a unique geodesic (at least in the
Yang-Mills and gravitational cases) and the fact that a geodesic cannot be simultaneously
orthogonal to and tangent to an orbit, one can show that V ultimately intersects all orbits.
To keep it from intersecting a given orbit more than once one may terminate each of the
generating geodesics as soon as it strikes a boundary point of /G. V is then topologically
(but not necessarily metrically) a copy of /G.
To gain an appreciation of some of the metrical situations that can arise think of as
being R
3
and G as being the group of screw motions with xed nonvanishing pitch about
some axis. The orbits are then helices and all, including the axis itself, are generic. If R
3
bears the Cartesian metric then the orbit space /G is topologically but not metrically a
plane. Note that in this example there exist no surfaces that intersect all orbits orthogonally,
although every plane not containing the axis is perpendicular to some orbit at its intersection
point and is a surface like V , based on that orbit.
Returning now to the general problem, we may place the identity element on each orbit
at the point where the orbit intersects V . If another subspace V

is constructed like V but

starting from another point on the base orbit, it too will intersect all the orbits. Because
4
These equations are readily veried to be integrable in virtue of the identities (29) and (62).
26
the group operations are isometries of
i

j
, the P

will be constants over V

. That is, once
the identity points are lined up all the other points are automatically lined up too. The
gauge condition (94) is therefore globally valid for all

in the domain G.
Unfortunately in practice it is almost hopelessly dicult to implement constructions like
this one, which are guaranteed to yield globally valid gauge conditions. In the present
construction, because V is generally orthogonal to none but the base orbit, one is faced
with the problem of solving global functional constraints rather than functional dierential
equations. Even the functional dierential equations that one has, namely Eqs. (95), are
highly nontrivial. In the Yang-Mills and gravitational cases they take respectively the forms
_
P

(x)/A

(x

)
_
;

= (

(P(x))(x, x

), (96)
2
_
P

(x)/g

(x

)
_
;

(P(x), x

), (97)
(see eqs. (22), (28), (54) and (55)).
By far the bulk of all work on Yang-Mills theory and quantum gravity has made use of
linear gauge conditions, i.e., conditions (94) with P

taken in the form

P

[] = P

i
[
B
]
i
,
i
=
i

i
B
(98)
where
i
B
is some ducial eld, often called a background eld. To ensure that the
subspaces dened by (94) do indeed intersect the orbits uniquely, at least in the vicinity of
the background eld and with the

close to zero, one often makes use of the orthogonality

idea by choosing
P

i
[
B
] = (1)
(j+1)
Q
i

[
B
]
j

i
[
B
]. (99)
For example, if
i

j
has the form (43) then, with the choice (99), the condition P

= 0
becomes
g
1/2
B
_
2

_
;
= 0,

= g

g
B
. (100)
Here indices are raised and lowered by means of the background metric g
B
and the covariant
derivative is dened in terms of it. With set equal to 1 (see the comments following eqs.
(46) and (47)) this is a very popular gauge condition in quantum gravity. The corresponding
condition in Yang-Mills theory, with
i

j
given by (45), is

;
= 0, (101)
27

= A

A

B
. (102)
Here indices are raised and lowered by means of the metrics

and

and the covariant

derivative is dened in terms of the background eld A

B
. Condition (101) is known as the
Lorenz condition.
Linear gauge conditions are extremely convenient in perturbation theory, where the eld

i
is treated as if it never gets very far from the background
i
B
. Covariant (with respect
to the background) gauge conditions like (100) and (101) are usually the best, but for some
purposes noncovariant gauges (e.g., the Coulomb gauge in Yang-Mills theory) are more
useful. In non-perturbative studies, however, linear gauge conditions have to be used with
great care (see Gribov (1977)). At least ve things can go wrong with linear gauge conditions
when applied globally:
(1) The subspace dened by a linear condition may or may not have a boundary, and if it
does this boundary may not coincide with the boundary (if any) of /G.
(2) The subspace dened by a linear condition may intersect some orbits more than once.
(3) There may be some orbits that it does not intersect at all.
(4) Even if it intersects all orbits when the

in eq. (94) have certain values, it may not

intersect all orbits when the

have other values. This means that there is no natural

domain for the

.
(5) When G is twisted there are no globally valid gauge conditions at all, linear or other-
wise.
If any of the above situations hold, the subspace dened by (94) will not represent /G
faithfully. It is possible in some cases to patch things up so that the advantages of linear
gauge conditions can be maintained. This has been done in certain global studies in Yang-
Mills theory. However, the dieomorphism group, as we have repeatedly emphasized, is a
much more complicated group than the Yang-Mills group and both the diculties to which it
gives rise globally and the opportunities that it presents for technical innovation are almost
unknown at the present time. In order to keep all options open we shall rst develop the
formal theory using foolproof gauge conditions, such as those based on group coordinates,
and then make some remarks about how things might go when other gauge conditions are
used.
28
XI. THE ACTION. VERTEX FUNCTIONS. RENORMALIZABILITY.
The dynamical behavior of any eld is determined by its action functional S. The action
functionals of (pure) Yang-Mills and gravity theories are respectively
S
A
=
1
4
_
F

d
n
x, (103)
S
g
= 2
_
g
1/2
Rd
n
x. (104)
As long as the limits of integration are not specied these integrals must be regarded as
purely formal expressions that serve merely to yield the dynamical equations:
0 = S
A
/A

F

;
, (105)
0 = S
g
/g

2g
1/2
_
R

1
2
g

R
_
. (106)
For some purposes, however, values need to be assigned to the actions. Integration bound-
aries must then be specied and, in the case of the gravitational eld, a surface integral must
be split o from (104) so that the integrand involves derivatives of g

of order no higher
than the rst.
We refer the student to standard references (e.g., Misner, Thorne and Wheeler (1973))
for analyses of the initial value problems associated with eqs. (105) and (106). From these
analyses it is readily deduced that in a spacetime of n dimensions the Yang-Mills eld has
n 2 degrees of freedom per spatial point and the gravitational eld has
1
2
n(n 3).
In the generic notation, eqs. (103) and (104) are written
S
,i
= 0. (107)
Gauge invariance of the theory is guaranteed by the identity
S
,i
Q
i

= 0, (108)
or, more explicitly,
F

;
= 0, 4
_
g
1/2
_
R

1
2
g

R
__
;
= 0. (109)
The left hand side of eq. (107) transforms linearly under the gauge group and hence the
gauge group leaves the eld equations intact. This is most easily seen by functionally
dierentiating eq. (108), which yields
S
,i
= S
,ij

j
= S
,ij
Q
j

= (1)
i
S
,j
Q
j
,i

. (110)
29
The action functionals (103), (104) may be expanded in functional Taylor series about a
background eld. In generic notation one writes
S = S
B
+ (S
,i
)
B

i
+
1
2!
(S
,ij
)
B

i
+
1
3!
(S
,ijk
)
B

i
+ ..., (111)

i
=
i

i
B
.
If the background elds satisfy the classical eld equations then the second term on the right
may be omitted.
The functional derivatives (S
,i
1
...i
N
)
B
with N 3 are known as (bare) vertex functions.
In the case of the Yang-Mills eld the vertex functions with N > 4 vanish and the Taylor
series terminates. In the case of the gravitational eld the series may or may not terminate
depending on what choice is made for the basic eld variables. By expressing inverse matrices
in terms of minors and determinants, and by examining the number of determinants needed
to yield unit total weight for the integrand of (104), one easily veries that if the basic
eld variables are taken to be (

g
r
g

and r is chosen to be 5/(4n + 2) then the vertex

functions with N > 2n+1 vanish. Alternatively, if (

= g
r
g

are chosen as the basic eld

variables, with r = 5/(6n2), then the vertex functions with N > 3n1 vanish.
5
There are
three reasons, however, why neither of these choices is useful. First, the vertex functions of
gravity theory are exceedingly complicated, involving thousands of terms already for N = 4.
Nobody is going to work out the vertex functions up to maximum order even with the aid
of a computer. Second, any imagined advantage in these choices is lost as soon as one tries
to introduce dimensional regularization into the quantum theory. A specic choice of eld
variables has then to be made, and it cannot vary continuously with the dimension. Third,
although a series that terminates has an innite radius of convergence, the range of the
variables

B
or

(
B
is in fact limited. These variables must avoid
regions where the signature of the metric tensor changes.
The third reason is the most important, at least in perturbation theory. As is well
known, the Feynman rules are obtained by inserting the expansion (111) into the Feynman
functional integral (see the next section) and evaluating the integral as a sum (asymptotic
series) of Gaussian integrals, with the
i
ranging from to . Any constraint on the
i
would make these integrals almost impossible to evaluate, and although one may for some
5
Both of these choices require n ,= 2. (See eqs. (40)).
30
purpose wish to extend the Feynman integrand into nonphysical regions, one never does this
by naively removing constraints.
These remarks suggest, in fact, that none of the variables (40) is good to use in pertur-
bation theory. A better choice would be something like

_
ln(g
1
)
_
, g = e

1
, (

), g (g

), (

),
1
(

), (112)
which maintains the signature of the spacetime metric. With these variables the series (111),
of course, does not terminate, and one speaks of gravity theory as being a non-polynomial
Lagrangian theory (Isham, Strathdee and Salam (1971),(1972)). It will be noted that all
such safe variables inevitably transform nonlinearly under the dieomorphism group.
Regardless of the choice of variables it is not dicult to draw preliminary conclusions
about the renormalizability or nonrenormalizability (in perturbation theory) of a given quan-
tum eld theory. Although momentum space is not, in an absolute sense, appropriate for use
in quantum gravity, conclusions about the high energy behaviour of amplitudes in perturba-
tion theory may be safely drawn with its aid. Consider a Feynman graph with L
e
external
lines, L
i
internal lines, and V
N
Nth-order vertices (N 3). L
e
, L
i
and V
N
are related by
the topological condition
L
e
+ 2L
i
=

N
NV
N
. (113)
The number of independent closed loops, or momentum integrations, in the graph is given
by
I = L
i

N
V
N
+ 1. (114)
In quantum gravity 2 powers of momentum are associated with each vertex, 2 powers with
each internal line, and n powers with each momentum integration. The supercial degree
of divergence of the graph is therefore
D = 2L
i
+ 2

N
V
N
+ nI = (n 2)I + 2, (115)
which, for n > 2, increases without limit as the number of independent closed loops increases.
This means that for n > 2, there is an innite number of primitive divergences, and, if
one attempts to compute order by order, an innite number of experimentally determined
coupling constants is needed to determine the theory. These conclusions are not altered if
account is taken of the ghost contributions which, as we shall see in the following sections,
must be included. The theory is said to be nonrenormalizable.
31
In Yang-Mills theory, in contrast, only one power of momentum is associated with each
3rd-order vertex, and the 4th-order vertices have no momentum dependence at all. This
leads to
D = 2L
i
+

N
(4 N)V
N
+ nI = 4 + (n 4)I L
e
. (116)
For n > 4 this theory too is nonrenormalizable, but for n = 4 and an arbitrary background
eld there are only four primitive divergences (corresponding to L = 1, 2, 3, 4), and the
theory is renormalizable. The proof of renormalizability is not trivial and depends crucially
on gauge invariance as well as some of the formal developments to be discussed in the
following sections. The primitive divergences turn out to be related in virtue of gauge
invariance.
The nonrenormalizability of standard quantum gravity has stimulated investigations of
alternative theories in which terms of the form g
1/2
_
R
2
+ R

_
are added to the
integrand of expression (104). Such theories generally suer from physical ghosts with
negative probabilities, but they do improve the convergence situation. Each vertex now
carries 4 powers of momentum and each internal line carries 4 powers. This leads to
D = 4L
i
+ 4

N
V
N
+ nI = (n 4)I + 4. (117)
When n = 4 all diagrams have the same supercial degree of divergence, namely 4. There
is an innity of primitive divergences, but they are all related by gauge invariance, and only
three experimental coupling constants are required. The proof of renormalizability has been
carried out by Stelle (1977) using methods similar to those applied to Yang-Mills theory.
It will be observed, in virtue of eq. (115), that the same methods should work in the case
of standard quantum gravity when n = 2, although since the number of degrees of freedom
in the eld is then negative it is not clear what such a theory means. Weinberg (1979) has
studied the asymptotic stability of quantum gravity when n = 2 + , << 1, and has given
plausibility arguments concerning its relevance for trying to make sense out of the theory
when n = 4.
In addition to its ultraviolet divergences quantum gravity also possesses infrared diver-
gences. Gravitational eld quanta -gravitons- are massless. This fact in itself need not
lead to diculties worse than those encountered in quantum electrodynamics where the di-
vergences are completely understood and are removable by standard methods. Gravitons,
32
however, are coupled to other massless quanta (photons, neutrinos, etc.) as well as to them-
selves. In Yang-Mills theory as well as in massless electrodynamics such a situation gives
rise to infrared divergences of a new type that cannot be removed by standard techniques
or argued away on physical grounds. In quantum gravity these new divergences are miracu-
lously absent (Weinberg (1965) and DeWitt (1967c)). It appears therefore that the mysteries
of Yang-Mills theory and gravity theory lie at opposite ends of the momentum spectrum.
There is an increasing body of evidence that the Yang-Mills eld solves its infrared dilemma
by adopting a nonstandard behaviour at long wavelengths, which is intimately related to
the phenomena of quark connement and dynamical symmetry breaking. These phenomena
may also bear a technical relation to the failure of gauge conditions to be globally valid in
Yang-Mills theory (Gtibov (1977)). No analogous phenomena are known to exist for gravity,
at least when spacetime is dieomorphic to R
n
. The mysteries of gravitation theory thus
appear to lie solely at the high end of the momentum spectrum.
XII. THE FEYNMAN FUNCTIONAL INTEGRAL. FACTORING OUT THE
GAUGE GROUP.
Consider a transition amplitude of the form out[in) where the vectors [in) and [out) refer
to states in which the eld is maximally specied (in the quantum mechanical sense, e.g.,
in terms of complete sets of commuting observables) in regions in and out respectively.
These states need not be vacuum states and the regions in and out need not refer to
the innite past and future respectively. If the background eld (which enters naturally in
most calculations) has singularities in the past and/or future, [in) and [out) may be dened
not in terms of observables at all but by some analytic continuation procedure (e.g., to the
Euclidean sector) that removes the singularities. It will be assumed only that the in
and out regions lie respectively to the past and future of the region of dynamical interest.
There are many ways of showing that the amplitude out[in) can be expressed as a formal
functional integral:
out[in) = N
_
e
iS[]
[]d, d

i
d
i
. (118)
Here N is a normalization constant, S[] is the classical action functional, [] is chosen
to make the volume element d gauge invariant (see eq. (50)), and the integration is to
be extended over all elds that satisfy the boundary conditions appropriate to the given
33
in and out states. We have remarked earlier (and will show later) that may be set
eectively equal to unity if the
i
are chosen to transform linearly under the gauge group.
We shall assume that such a choice has been made and henceforth drop from the theory.
(It can always be restored if desired.)
Expression (118) was rst derived by Feynman (1948) in ordinary quantum mechanics,
without gauge groups, and later (1950) applied by him to eld theory. The full extension
to eld theories with gauge groups is the work of many people, and the student is referred
to the literature for details.
6
When the Feynman integral is applied to the gravitational
eld the only additional comment that needs to be made is that the integration may have
to embrace as many topologies as can be reached by analytic continuation from the given
background topology.
If any of the elds
i
in (118) are fermionic the integration with respect to them is to be
carried out according to the formal rules for integrating with respect to anticommuting vari-
ables that were rst introduced by Berezin (1966). These rules are analogous in many ways
to the well known rules for ordinary denite integrals from to with integrands that
vanish asymptotically. For example, integrals of total derivatives vanish, and the position
of the zero point may be shifted. On the other hand, with Berezin rules, transformations of
variables and evaluation of Gaussian integrals lead to determinants precisely inverse to those
of standard theory. When both bosonic and fermionic elds are involved it is the super
determinant that appears.
All physical amplitudes can be deduced from expression (118) by examining how out[in)
changes under variations in the action. Physical amplitudes can alternatively be obtained
by judicious use of
out[T(A[])[in) = N
_
A[]e
iS[]
d, (119)
where A[] is any functional of the eld operators
i
, and the T symbol removes ambiguities
about ordering the
i
by arranging them chronologically (with appropriate signs thrown
in if any of the
i
are fermionic).
When a gauge group is present the integration in (118) is redundant. Furthermore (119)
is generally ambiguous, unless A is gauge invariant, in which case the integration in (119)
too is redundant. This is because, owing to the gauge invariance of the classical action, the
6
Useful modern references are Faddeev (1969), (1976) and Abers and Lee (1973).
34
exponent in the integrands of (118) and (119) remains constant as ranges over a group
orbit in the conguration space . One can remove this redundancy and/or ambiguity
by adopting a gauge condition like (94). The details of the procedure were rst given by
Faddeev and Popov (1967).
Let be an element of the gauge group G, with coordinates

, and let

be the eld
to which is displaced under the action of . Dene
[, ]
_
G
[P[

] ]detQ
1
[]d, d

, (120)
where [ ] is the delta functional, P

are the functionals appearing in eq. (94), Q

1
is the
inverse of the matrix formed out of the Q

of eq. (53), and the integration extends over the

entire gauge group! We shall assume that the gauge condition (94) is globally valid. The
integrand in (120) then switches on at only one point in G, namely that point for which

is equal to the unique eld

lying on the orbit containing and picked out by the

gauge condition:
P

] =

. (121)
By building innitesimal parallelepipeds in the group manifold G and making use of eq.
(53) one can verify that the combination detQ
1
[]d appearing in (120) is a right-invariant
volume element, satisfying
detQ
1
[

]d(

) = detQ
1
[]d for all

in G. (122)
The presence of this volume element renders the functional gauge invariant:
[,

] = [, ] for all

in G. (123)
Several comments must be made about its use, however. In the case of the dieomorphism
group, with the Qs given by (54), it is easily checked that
Q
1

[] =

(x, (x

))
((x

))
(x

)
. (124)
No one has ever discovered how to evaluate or give a meaning to the determinant of this
continuous matrix. The right-invariant volume element of the dieomorphism group, there-
fore, can only be dened (and used) purely formally. The same is true for the invariance
group of supergravity theory. It should be noted that when the group is a supergauge group,
35
possessing anticommuting as well as commuting coordinates, det Q
1
is a superdeterminant,
and the integral (120) involves the Berezin rules. (Remark: the delta functional in (120)
presents no diculty. Delta functions of anticommuting variables turn out to be easy to
dene. They can even be given Fourier representations.)
The gauge invariance of makes it an easy functional to evaluate. One has only to shift
to

so that the integrand in (120) switches on at the identity element I. All quantities
can then be expanded in power series in

. For example, the argument of the delta

functional becomes
P

= P

+ P

,i
[

]Q
i

](

) + ...
= F

](

+ I

) + ... (125)
where
F

[] P

,i
[]Q
i

[]. (126)
Similarly, making use of eq. (68), we nd
Q
1

[] =

,
[I](

) + ... (127)
and hence
[, ] =
_
G

_
F[

]( I) + ...
__
1 (1)

,
[I](

) + ...
_
d
=
_
detF[

]
_
1
, (128)
F being the matrix with elements F

. If any of the group indices is fermionic the determinant

is again a superdeterminant. Note that if the Ps are constructed according to eq. (95),
F[] is identical to the matrix Q[P[]]. Although this construction will not be assumed in
what follows, we shall, for simplicity and convenience, assume that the gauge condition (94)
is globally valid for all

lying in the ranges of the functionals P

[].
The next step is to insert unity into the integrand of (118), in the guise of
([, ])
1
_
G

_
P[

]
_
detQ
1
[]d,
and interchange the order of integrations, obtaining
out[in) = N
_
G
detQ
1
[]d
_
de
iS[]
([, ])
1

_
P[

]
_
. (129)
36
We have assumed a choice of variables for which the volume element d is gauge invariant
( = 1). S[] and [, ] are also gauge invariant. Therefore a superscript may be axed
to every in the integrand of (129) that does not already bear one. But every

is then
a dummy, and hence all the s may be removed. Making use of (128) one immediately
obtains
out[in) = N

_
e
iS[]
detF[]
_
P[]
_
, (130)
where F[

] has been replaced by F[] in the integrand because of the presence of the delta
functional, and where
N

N
_
G
detQ
1
[]d. (131)
The gauge group has now been factored out, and its volume has been absorbed into
the new normalization constant N

. The integration in (130) is restricted to the subspace

P

[] =

.
The technique of conning the elds
i
to a particular subspace can also be used to
remove the ambiguity from the integral (119) when A[] is not gauge invariant. Strictly
speaking, matrix elements are denable only for gauge invariant operators. However, given
a non-gauge-invariant operator A[], one can construct a gauge invariant operator out of it
by the following denition:
T(A[

]) T
_
([, ])
1
_
G
A[

]
_
P[

]
_
detQ
1
[]d.
_
(132)
The chronological ordering symbol is used here so that the non-commutativity (or anti-
commutativity) of A[

] with both ([, ])

1
and the delta functional can be eectively
ignored. Note that because the gauge group acts linearly on the s there is no ambiguity
about the symbol

. Note, however, that dieomorphisms in gravity theory can drag the
eld in very complicated ways. The chronological operation, which orders eld operators
solely by the value of the coordinate x
0
, rearranges the physical elds in correspondingly
complicated ways as the variable in the integral (132) ranges over the group.
Applying eq. (119) to the operator T(A[

]) and following the same reasoning as was

used in passing from eq. (129) to eq. (130), one nds
out[T(A[

])[in) = N

_
A[]e
iS[]
detF[][P[] ]d, (133)
valid for any functional A[].
37
XIII. AVERAGING OVER GAUGES
It is possible to develop a perturbation theory based on eqs. (130) and (133), but it is
usually more convenient to work with a formalism from which the delta functionals have
been eliminated. Note that although the parameters

appear on the right side of (130),

the amplitude out[in) is actually independent of them. Therefore nothing changes if we
integrate over these parameters, with a weight factor.
In practically all studies of non-Abelian gauge theories to date, Gaussian weight factors
of the form
exp
_
1
2
i

_
,
where M is a nonsingular constant matrix having the symmetry

= (1)
++

,
have been used. From a fundamental standpoint a Gaussian weight factor can be used only
if the bosonic s can range from to without the gauge condition (94) becoming
globally invalid - for example, if the Ps satisfy eq. (95), with Q

s based on canonical
group coordinates (eq. (69)). This condition, however, is almost universally violated. In-
deed, the Gaussian weight function is most frequently employed in combination with linear
gauge conditions like (98), where it almost certainly introduces errors globally (i.e., in non-
perturbative analyses).
In the case of quantum gravity, where the gauge group has no canonical coordinates, it
seems particularly inappropriate to conne our attention to Gaussian weight factors. We
shall therefore introduce a more general weight factor, of the form exp(iU[]), where we
specify nothing about the functional U[] except the following three conditions:
(1) U becomes innite on the boundary of the allowable domain of the s (which domain
we are assuming coincides with the range of the P[]s).
(2) U and all its rst functional derivatives U
,
vanish at some chosen point (e.g., at

= I

when the Ps are group coordinates); its second functional derivatives U

,
, however, form
a nonsingular continuous matrix at that point.
(3) U vanishes nowhere else, and its rst derivatives all vanish simultaneously nowhere else.
38
The third condition is imposed mainly for convenience. Note that all three are satised
by the Gaussian exponent
1
2

whenever it can be legitimately used.

Inserting exp(iU[]) into the integrand of eq. (130) and integrating over the s, one
obtains
out[in) = N

[U]
_
e
i(S[]+U[P[]])
detF[]d, (134)
N

[U]
N

_
e
iU[]
d
, (135)
the integration domain in (135) being understood to be the allowable domain of the s.
Equation (133) too may be replaced by a weighted average. Dening
T(A[])
_
_
e
iU[]
d
_
1
_
T(A[

])e
iU[]
d, (136)
one may write
out[T(A[])[in) = N

[U]
_
A[]e
i(S[]+U[P[]])
detF[]d. (137)
Equation (137), and generalizations of it, will be used frequently in the following sections.
Denitions (132) and (136) reveal precisely what kind of averaged quantum operator is
associated with each classical functional A[] in this formalism. Note that if f[] is any
functional of the s we have
T(f[P[

]]) = f[] (138)

and
out[T(f[P[]])[in) = f)out[in) (139)
where
f)
_
f[]e
iU[]
d
_
e
iU[]
d
. (140)
Having so freely manipulated formal expressions we should now check that no inconsis-
tencies have crept into our results, by verifying directly that the right side of (134), for
example, is truly independent of the choices we have made for the functionals P

[] and
U[]. Obviously the right side will be aected if we naively switch to Ps for which the
gauge condition (94) is no longer globally valid for all s in the range of the Ps. Therefore
we must assume that the changes P

(which, without loss of generality, may be taken

innitesimal) maintain global validity.
39
We also conne our attention to changes U that leave the location of the zero of U, as
well as the three conditions that we imposed upon U, intact. It is not dicult to see that
U may then always be expressed in the form
U[] = U
,
[]V

[], (141)
where the V

vanish at the zero of U. Note that under this change we have

N

[U] = iN

[U]
_
e
iU[]
U
,
[]V

[]d
_
e
iU[]
d
= N

[U](1)

,
), (142)
the nal form being obtained by an integration by parts in which the boundary of the
integration domain contributes nothing because exp(iU[]) oscillates innitely rapidly there.
Making use of eqs. (126), (139), (141) and (142) we now have
out[in) = N

[U]
_
e
i(S[]+U[P[]])
_
(1)

,
[P[]]
+ iU
,
[P[]]
_
V

[P[]] + P

[]
_
+ (1)

F
1

[]P

,i
[]Q
i

[]
_
detF[]d, (143)
where the inverse F
1
, if it is a Greens function (as it often will be), must satisfy the
boundary conditions appropriate to the in and out states. The integral (143) does not
obviously vanish. The way to show that it is nevertheless zero is as follows. Replace each

i
in the integral (134) by
i
, where

i
=
i
+ Q
i

[]

[], (144)

[] = F
1

[]
_
V

[P[]] + P

[]
_
. (145)
Since the s are just dummies this replacement has no eect. However, it is not dicult to
show that the net apparent change in the integral is given precisely by (143) provided one
is entitled to make the identications
(1)
i(+1)
Q
i
,i
= 0, (1)
(+1)
C

= 0. (146)
We shall comment on these equations presently.
It is easy to see that the second term inside the curly brackets in (143) comes from the
change that the replacement induces in the exponent of (134). That the rst and
40
third terms come from the change in the product detF[]d may be shown as follows. First
compute

i
,j
=
i
j
+ (1)
j
Q
i
,j
[]

[]
Q
i

[]F
1

[]
_
(1)
jk
P

,kj
[]Q
k

[] + (1)
j
P

,k
[]Q
k
,j
[]
_

[]
+Q
i

[]F
1

[]
_
V

,
[P[]]P

,j
[] + P

,j
[]
_
,
which, after rearrangement of some factors and use of eq. (126), yields the (super) Jacobian
det
_

i
,j
_
= 1 + (1)
i(+1)
Q
i
,i
[]

[]
F
1

[]
_
(1)
(j+1)
P

,ji
[]Q
i

[]Q
j

[]
+ (1)
(+1)
P

,j
[]Q
j
,i
[]Q
i

[]
_

()
+ (1)

V

,
[P[]] + (1)

F
1

[]P

,i
[]Q
i

[]. (147)
Combining d d d =
_
det
_

i
,j
_
1
_
with
detF[ detF[] detF[]
= (1)

detF[]F
1

[]F

,i
[]Q
i

[]

[]
= (1)

detF[]F
1

[]
_
(1)
i
P

,ij
[]Q
j

[]Q
i

[]
+P

,j
[]Q
j
,i
[]Q
i

[]
_

[],
and making use of eq. (29), one nds for the change in detF[]d under the replacement
,

_
detF[]d
_
=
_
_
(1)
i(+1)
Q
i
,i
(1)
(+1)
C

[]
+ (1)

,
[P[]] + (1)

F
1

[]P

,i
[]Q
i

[]
_
detF[]d. (148)
If eqs. (146) are assumed to hold one is left with precisely the rst and third terms inside
the curly brackets in (143).
Equations (146) were not needed in the derivation of eq. (134). Why are they needed
now? When the gauge group has no anticommuting coordinates the answer is that eqs.
(146) are forced on us by the procedure of factoring out the gauge group. Our interchanging
the orders of integration in arriving at eq. (129), and our use of eq. (131), amount to
41
adopting the rule that the gauge group is to be treated formally as if it were compact. For
consistency the associated Lie algebra must likewise be treated as compact. The generators
of real representations of compact Lie algebras all have vanishing trace. Hence eqs. (146).
(Remember, we are assuming that the s transform linearly under the gauge group.)
In Yang-Mills theories eqs. (146) hold automatically because the generating group is
always compact. In gravity theory the situation is more subtle. Both Q
i
,i
and C

, if one
tries to compute them from eqs. (28) and (35), are meaningless expressions involving deriva-
tives of delta functions with coincident arguments. However, both are metric-independent
covariant vector densities of unit weight. Any sensible regularization scheme must assign
them the value zero, for otherwise spacetime would be endowed with a preferred direction
even before a metric is imposed on it.
If G is a supergauge group, with anticommuting coordinates, the formal compactness
argument fails. But the notion of simplicity, or semisimplicity, survives. The invariance
groups of all known supergauge theories are semisimple, and the generators of real represen-
tations of such groups satisfy the supertrace laws (146). The semisimplicity argument can
also be invoked in the case of the local frame group, which enters when the gravitational
eld is expressed in terms of local frame components rather than directly in terms of the
metric tensor (e.g., when spinor elds are present).
If we choose s that do not transform linearly under the group then the functional []
of eq. (118) has to be reintroduced into the theory. It is easy to verify that consistency
of the above formalism is maintained under these circumstances provided the rst of eqs.
(146) is replaced by eq. (50), which is just the condition that the product []d be gauge
invariant. Equation (50) is, of course, consistent with the rst of eqs. (146) when = 1.
At this point the student may object that, in the case of quantum gravity at least, there
appears to be an inconsistency in what we have done. Consider the sets of variables dened
by eqs. (40). All of these sets transform linearly under the dieomorphism group. However,
the Jacobian that arises in transforming from one set to another is not generally constant.
How can one maintain = constant for all sets? The answer is that one must. If the
Jacobian is replaced by the exponential of its logarithm, it contributes a formally divergent
term of the form const. (0)
_
lngd
n
x to the exponents in the Feynman functional integrals.
All terms of this kind must be suppressed by any viable regularization scheme. By this
criterion the dimensional regularization method, for example, is a viable scheme.
42
XIV. GHOSTS. THE BRS TRANSFORMATION. THE GENERATING FUNC-
TIONAL
The perturbation rules to which eqs. (134) and (137) lead may be summarized as follows.
The exponent in the integrands is, as usual, expanded about a stationary background
B
,
and the integrals are evaluated as series of Gaussian integrals. If exp(iU) is a Gaussian
weight factor and the P

are chosen (unwisely) to have the linear form (98), then the vertex
functions are just the functional derivatives S
,i
1
...i
N
, N 3. Otherwise the vertex functions
include contributions from U[P[]]. In addition to the usual graphs that one can draw there
is an innite set of new graphs arising from the factor detF[], involving a new set of formal
particles called ghosts. The inverse matrix F
1

[] is the bare ghost propagator in an

arbitrary eld , i.e., with an arbitrary number of -lines attached. The ghost propagators
always enter in closed loops, never as external lines.
The conditions previously imposed on the functional U ensure that the -propagator
exists. The presence of U[P[]] in the exponents of expressions (134) and (137) breaks the
gauge symmetry and eliminates the redundancy that exists in the integration (118). An
important symmetry nevertheless survives. It is most easily revealed by introducing two
new elds,

and

, that have the unusual property of being fermionic when the index
is bosonic and vice versa. Use of these elds together with the Berezin integration rules
allows one to express detF[] in the form
_
e
iF

[]

d d = C detF[], (149)
where C is a (divergent) constant, and hence
out[in) = N[U]
_
e
i(S[]+U[P[]]+F[])
d d d (150)
N[U] N

[U]/C. (151)
Equation (150) shows that the elds

are associated with the ghost particles, which

are now placed on a common footing with the -particles.
It was discovered by Becchi, Rouet and Stora (BRS) (1975) that both the exponent and
the volume element d d d in (150) are invariant under a set of transformations whose
innitesimal forms are given by
d
i
= Q
i

[]

= U
,
[P[]],

=
1
2
C

, (152)
43
where is an arbitrary innitesimal anticommuting constant. Using the special
(anti)commutativity properties of the s and s, together with the identity (29) and the
denition (31), one readily veries the invariance of the exponent. By computing the super-
Jacobian of the BRS transformation one nds that the volume element d d d is likewise
invariant, provided eqs. (146) are assumed to hold. It is also straightforward to verify
that, if conned to the s and s, the BRS transformations constitute an Abelian group.
Inclusion of the s destroys the group property unless F

[]

= 0. Note that the BRS

transformations do not constitute a local gauge group. The s are constants; they are not
functions over spacetime. Thus the integral (150) contains no redundancy.
The BRS transformations play an important role in simplifying the derivation of the
Ward-Takahashi identity satised by the so-called generating functional. What follows is
a partial account, adapted to the case in which the Ps are nonlinear and exp(iU) is non-
Gaussian, of the theory of the generating functional given by B.W. Lee (1976) and originally
due to Zinn-Justin.
One begins by replacing the exponent in eq. (150) by

S[, , , K, L, M] + J
i

i
+ J

+

J

,
where

S[, , , K, L, M] S[] + U[P[]] +

[]

+ K
i
+ M(U[P[]])
,i
Q
i

[]

1
2
(1)

(153)
and by generalizing eq. (137) to
out[T(A[, , ])[in) N[U]
_
A[, , ]e
i(

S+J+J+

J)
d d d. (154)
J
i
, J

,

J

, K
i
, L

and M are external sources

7
, and the matrix element (154) is a func-
tional of them. J
i
and L

are bosonic when their indices are bosonic and fermionic when
their indices are fermionic. With J

,

J

and K
i
the association is just the opposite. M is
fermionic.
If the functional A in (154) is replaced by unity one gets a generalization of the in-out
amplitude:
e
iW[J,J,

J,K,L,M]
out[in)
7
M is a constant. The others depend, through their indices, on position in spacetime.
44
N[U]
_
e
i(

S+J+J+

J)
d d d. (155)
This generalized amplitude is called the generating functional, because if it is expanded
in a power series in the sources J
i
, J

and

J

the coecients are the matrix elements of

chronological products of eld operators. The coecient of zero order reduces to the original
amplitude (150) when K
i
, L

and M vanish.
The functional

S may be viewed as a generalized action functional. With the aid of
eqs. (29) and (36) one may readily show that it is BRS invariant. Suppose the variables
, , in the integrand of (155), as well as in the volume element, are subjected to a
BRS transformation. Since these variables are dummies the integral remains unaected.
Explicitly, however, the terms in J, J,

J change. Therefore
0 = iN[U]
_ _
J
i
Q
i

[]

+ (1)

U
,
[P[]] +
1
2
(1)

_
e
i(

S+J+J+

J)
d d d. (156)
This result can be expressed in an alternative form through use of
0 =
_

_
f[]e
i(

S+J+J+

J)
_
d d d
= i
_
_
F

[]

(1)

_
f[]e
i(

S+J+J+

J)
d d d, (157)
where f is any functional of the s. One obtains
0 = iN[U]
_ _
J
i
Q
i

[]

+ U
,
[P[]]F

[]

+
1
2
(1)

_
e
i(

S+J+J+

J)
d d d
= N[U]
_
_
J
i

K
i
+

M


J

_
e
i(

S+J+J+

J)
d d d, (158)
in which use is made of the fact that the term containing M in

S may be written in the
form
MU
,
[P[]]F

[]

.
Multiplying eq. (158) by ie
iW
, we nally get
0 = ie
iW
_
J
i

K
i
+

M


J

_
e
iW
= J
i
W
K
i
+
W
M


J

W
L

. (159)
45
This relation expresses an important symmetry property of the generating functional, which
leads directly to the Ward-Takahashi identities to be discussed presently. First we must
review some standard material on the so-called eective action.
XV. MANY-PARTICLE GREENS FUNCTIONS. THE EFFECTIVE ACTION
In this section important use will be made of the Schwinger average:
A)
out[T(A)[in)
out[in)
. (160)
Here A is an arbitrary functional of the operators
i
,

, and the numerator and denom-

inator on the right are dened by eqs. (154) and (155) respectively. It will be convenient to
dene

i

i
),

),

). (161)
When the sources J

,

J

, K
i
, L

, M vanish, the averages

and

vanish. Note that

although the symbols
i
,

have previously been used for integration variables, no

confusion about their meaning will arise in practice.
It will also be convenient to denote the operators
i
,

, and

collectively by
A
, their
averages by
A
, and the sources J
i
, J

and

J

collectively by J
A
. Let J
A
be arbitrary
nite increments in the sources. Then we may write

n=0
i
n
n!
J
An
... J
A
1
out[T(
A
1
...
An
)[in)
= exp
_
J
A

J
A
_
out[in) =
_
e
iW
_
JJ+J
= exp
_
iW + i J
A

A
+ i

n=2
1
n!
J
An
... J
A
1
G
A
1
...An
_
, (162)
where

A
=
A
) = e
iW

iJ
A
e
iW
=
W
J
A
, (163)
G
A
1
...An

J
A
1
...

J
An
W. (164)
Dividing both sides of eq. (162) by e
iW
and comparing like powers of J
A
, one obtains an
innite sequence of relations:

B
) =
A

B
iG
AB
,
A

C
) =
A

C
iP
3

A
G
BC
+ (i)
2
G
ABC
, etc., (165)
46
where P means sum over the N distinct permutations of indices, with a plus sign or a minus
sign according to whether the permutation of the indices associated with fermionic elds
is even or odd. G
AB
is known as the one-particle propagator, and the G
A
1
...An
, n 3
are known as many-particle Greens functions. They satisfy the boundary conditions
specied by the in and out states.
Any functional of the sources J
A
may be alternatively regarded as a functional of the
averages
A
. From equations (163) and (164) one sees that the one-particle propagator is
the transformation matrix from one set of variables to the other:
G
AB
=

B
J
A
. (166)
This fact may be used to establish an important relation between the functional W and the
Schwinger average of the operator eld equations. The latter is obtained from the formal
functional identity
0 = ie
iW
N[U]
_
e
i(

S+J+J+

J)

A
d d d
=

S
,A
) + J
A
, (167)
where

S
,A
is the operator corresponding to the functional

S
,A
.
If we dierentiate eq. (167) on the left with respect to J
B
and make use of (166) we
obtain
G
BC
C,

S
,A
) =
B
A
. (168)
Here the functional derivative inside the brackets ) is with respect to the eld operator

A
, and the functional derivative outside the brackets is with respect to the eld average

C
.
C,

S
,A
) is seen to be the operator of which the one-particle propagator G
BC
is the
Greens function. Because of its boundary conditions G
BC
may be shown to be both a left
Greens function, as in eq. (168), and a right Greens function of
A,

S
,B
) as well. It has the
symmetry
G
AB
= (1)
AB
G
BA
(169)
which implies
A,

S
,B
) = (1)
A+B+AB
B,

S
,A
). (170)
But this is just the condition that there exist a functional

[, , , K, L, M] such that

,A
=

S
,A
). (171)
47

is known as the eective action. It satises the equations

,A
= J
A
, (172)
A,

,C
G
CB
=
B
A
, (173)
and is related to the functional W by a Legendre transformation:
W =

+ J
A

A
. (174)
This relation nay be veried through dierentiation with respect to J
B
and use of eq. (172)
in the form
A,

= (1)
A
J
A
.
Thus
W
J
B
=

A
J
B
_
A,

+ (1)
A
J
A
_
+
B
=
B
,
which is just eq. (163). Since

is determined only up to an arbitrary constant of integration,
eq. (174) may be regarded as xing it.

is also known as the generating functional for proper vertices. This stems from its
relation to the many-particle Greens functions. By dierentiating eq. (173) one can relate
functional derivatives of the one-particle propagator to derivatives of

. These relations
yield, for example,
G
ABC
=

J
A
G
BC
= G
AD
D,
G
BC
= (1)
(B+C)D+(C+D)E+(D+E)F
G
AD
G
BE
G
CF
DEF,

. (175)
If the propagators are represented by lines and the third and higher derivatives of

are
represented by vertices, one easily sees that each new dierentiation with respect to a source
inserts a new line in all possible ways into the previous diagram. Each Greens function of
given order is thus representable as a sum of all the possible tree diagrams of that order.
Suppose the spatial sections of spacetime are noncompact. Then an S-matrix can be
introduced, connecting states dened at innity. The S-matrix is expressible in terms of
the chronological products appearing in eq. (162). Because these products are expressible
in terms of the Greens functions (eqs. (165)), it follows that when

is used, only tree
diagrams are needed in the construction of the S-matrix. No closed loops appear. The
vertices generated by

are the proper vertices, already containing all quantum corrections.
48
By noting that identical tree diagrams occur in classical perturbation theory, but with

replaced by

S, one can show that

describes the quantum-corrected dynamics of coherent
large-amplitude elds. One must expect the same to be true also when the spatial sections
are compact and there is no S-matrix.
XVI. THE WARD-TAKAHASHI IDENTITY
We now resume use of the symbols
i
,

, J
i
, J

,

J

and rewrite eq. (174) in the more

explicit form
W[J, J,

J, K, L, M] =

[, , , K, L, M] + J
i

i
+ J

+

J

. (176)
The averages
i
,

depend on all six sources, but because K

i
, L

and M do not partic-

ipate in the Legendre transformation one may show that
W
K
i
=

K
i
,
W
L

,
W
M
=

M
, (177)
where the derivatives on the right refer only to the explicit dependence of

on K
i
, L

and
M. This result, combined with eq. (172) in the form
A,

= (1)
A
J
A
,
allows eq. (159) to be rewritten as
(1)
i

K
i
+

M
(1)

= 0, (178)
all derivatives being left derivatives. This is the Ward-Takahashi identity.
The Ward-Takahashi identity has important implications for the structure of

. That
it implies the existence of some sort of symmetry possessed by

becomes obvious when
one notes that, because of its BRS invariance,

S too satises the Ward-Takahashi identity.
Unfortunately, to work from eq. (178) to the symmetry possessed by

is a much harder task.
In principle one might do the following. Assume that

can be expanded as a power series
in

, K
i
, L

and M. Such an assumption has nothing a priori to do with perturbation

theory since the expansion is to be carried out after the functional integration (155) has
been performed. It is based on the reasonable belief that (155) varies smoothly (at least
49
after appropriate renormalizations) as K
i
, L

, M, J

and

J

(and hence

and

) go to
zero.
In determining the kinds of terms that can appear in the expansion it is useful to introduce
the notion of ghost number. If one assigns the ghost numbers 1 to

and J

; 0 to
i
and
J
i
; 1 to

,

J

, K
i
and M; and 2 to L

; one easily sees that the integrand in (155) and the

integral itself have total ghost number zero. Hence W and

have total ghost number zero,
and all the terms in the expansion of

must have this property as well. Also the expansion
can contain no terms in M of higher order than the rst since M is an anticommuting
constant.
If one inserts the expansion into (178) and groups together terms of like powers, one ob-
tains an innite sequence of subsidiary Ward-Takahashi identities relating the -dependent
coecients. Unfortunately there seems to be no easy way of drawing simple inferences from
these identities en gros. So far (178) has been applied only to renormalizable models in
perturbation theory. There it has proved to be of great service in the practical details of the
renormalization program, as well as in the demonstration that the theory is indeed renor-
malizable to all orders and that unitarity is maintained. (The student is referred to the
literature for details.) What role it is destined to play in quantum gravity remains to be
seen.
XVII. REFERENCES IN THE DEWITT LECTURES
Becchi, C., A. Rouet and R. Stora (1975). Comm. Math. Phys. 42, 127.
Berezin, F.A. (1966), The Method of Second Quantization, Academic Press, New York.
DeWitt, B.S. (1965), Dynamical Theory of Groups and Fields, Gordon and Breach,
New York.
DeWitt, B.S. (1967a), Phys. Rev. 160, 1113.
DeWitt, B.S. (1967b), Phys. Rev. 162, 1195.
DeWitt, B.S. (1967c), Phys. Rev. 162, 1239.
Faddeev, L.D. (1976), in Methods in Field Theory, R. Balian and J. ZinnJustin, eds.,
50
1975 Les Houches Lectures, North Holland, Amsterdam.
Feynman, R.P. (1948), Rev. Mod. Phys. 20, 367.
Feynman, R.P. (1950), Phys. Rev. 80, 440.
Feynman, R.P. (1963), Acta Phys. Polon. 24, 697. See also in Proceedings of the 1962
Warsaw Conference on the Theory of Gravitation, PWN-Editions Scientiques de
Pologne, Warszawa (1964).
Fischer, A.E. (1970), in Relativity: Proceedings of the Relativity Conference in the
Midwest, Carmeli, Fickler and Witten, eds., Plenum, New York.
Freifeld, C. (1968), in Battelle Rencontres, 1967 Lectures in Mathematics and
Physics, C.M. DeWitt and J.A. Wheeler eds., W.A. Benjamin, Inc., New York.
Gribov, V.N. (1977), Lecture at the Twelfth Winter School of the Leningrad Nuclear Physics
Institute. (SLAC translation No. 176).
Isham, C.J., J. Strathdee and A. Salam (1971). Phys. Rev. D3, 1805.
Isham, C.J., J. Strathdee and A. Salam (1972). Phys. Rev. D5, 2584.
Kostant, B. (1977), in Dierential Geometrical Methods in Mathematical Physics
(Proceedings of the July 1-4, 1975 Symposium in Bonn), Bleuler and Reetz, eds. Lecture
Notes in Mathematics No. 570, Springer, Berlin.
Lee, B.W. (1976), in Methods in Field Theory, R. Balian and J. Zinn-Justin, eds., 1975
Les Houches Lectures, North Holland, Amsterdam.
Misner, C.W., K.S. Thorne and J.A. Wheeler (1973). Gravitation, Freeman, San Francisco.
Nath, P. (1976), in Proceedings of the Conference on Gauge Theories and Modern
Field Theory, Boston, 1975 (M.I.T. Press, Cambridge, Massachusetts).
Stelle, K.S. (1977), Phys. Rev. D16, 953.
Utiyama, R. (1956), Phys. Rev. 101, 1597.
Weinberg, S. (1965), Phys. Rev. 140, B516.
51
Weinberg, S. (1979), in General Relativity, an Einstein Centenary Survey, S.W.
Hawking and W. Israel, eds., Cambridge University Press.
XVIII. OLD UNIFICATION
We can now outline how the space-of-histories formulation of sections 2-16 provides a
common ground for describing the old and new unications of fundamental theories.
Quantum eld theory begins once an action functional S is given, since the rst and most
fundamental assumption of quantum theory is that every isolated dynamical system can
be described by a characteristic action functional [41]. The Feynman approach makes it
necessary to consider an innite-dimensional manifold such as the space of all eld histories

i
. On this space there exist (in the case of gauge theories) vector elds
Q

= Q
i

i
(179)
that leave the action invariant, i.e. (see eq. (108)),
Q

S = 0. (180)
The Lie brackets of these vector elds lead to a classication of all gauge theories known so
far.
A. Type-I gauge theories
The peculiar property of type-I gauge theories is that the Lie brackets [Q

, Q

] are equal
to linear combinations of the vector elds themselves, with structure constants, i.e.,
[Q

, Q

] = C

, (181)
where C

,i
= 0. The Maxwell, YangMills, Einstein theories are all example of type-I the-
ories (this is the unifying feature). All of them, being gauge theories, need supplementary
conditions, since the second functional derivative of S is not an invertible operator. After
imposing such conditions, the theories are ruled by a dierential operator of DAlembert
type (or Laplace type, if one deals instead with Euclidean eld theory), or a non-minimal
operator at the very worst (for arbitrary choices of gauge parameters). For example, when
52
Maxwell theory is quantized via functional integrals in the Lorenz [42] gauge
8
, one deals
with a gauge-xing functional
(A) =
b
A
b
, (182)
and the second-order dierential operator acting on the potential reads as
P
b
a
=
b
a
+ R
b
a
+
_
1
1

b
, (183)
where is an arbitrary gauge parameter. The Feynman choice = 1 leads to the minimal
operator

P
b
a
=
b
a
+ R
b
a
,
which is the standard wave operator on vectors in curved spacetime. Such operators play a
leading role in the one-loop expansion of the Euclidean eective action.
B. Type-II gauge theories
For type-II gauge theories, Lie brackets of vector elds Q

are as in eq. (181) for type-I

theories, but the structure constants are promoted to structure functions. An example is
given by simple supergravity (a supersymmetric gauge theory of gravity, with a symmetry
relating bosonic and fermionic elds) in four spacetime dimensions, with auxiliary elds [34].
C. Type-III gauge theories
In this case, the Lie bracket (181) is generalized by
[Q

, Q

] = C

+ U
i

S
,i
, (184)
and it therefore reduces to (181) only on the mass-shell, i.e., for those eld congurations
satisfying the EulerLagrange equations. An example is given by theories with gravitons and
gravitinos such as BoseFermi supermultiplets of both simple and extended supergravity in
any number of spacetime dimensions, without auxiliary elds [34].
8
In [42] the author, L. Lorenz, built a set of retarded potentials which can be shown to satisfy the Lorenz
gauge, although in 1867 no-one had thought of electrodynamics as a gauge theory. This author is not H.
Lorentz, whose name is incorrectly associated to such a gauge in previous literature.
53
D. From supergravity to general relativity
It should be stressed that general relativity is naturally related to supersymmetry, since
the requirement of gauge-invariant RaritaSchwinger equations [43] implies Ricci-atness in
four dimensions [44], which is then equivalent to vacuum Einstein equations. The Dirac
operator is more fundamental in this framework, since the m-dimensional spacetime metric
is entirely re-constructed from the -matrices, in that
g
ab
=
1
2m
tr(
a

b
+
b

a
). (185)
XIX. NEW UNIFICATION
In modern high energy physics, the emphasis is no longer on elds (sections of vector
bundles in classical eld theory, operator-valued distributions in quantum eld theory), but
rather on extended objects such as strings. In string theory, particles are not described
as points, but instead as strings, i.e., one-dimensional extended objects. While a point
particle sweeps out a one-dimensional worldline, the string sweeps out a worldsheet, i.e., a
two-dimensional real surface. For a free string, the topology of the worldsheet is a cylinder
in the case of a closed string, or a sheet for an open string. It is assumed that dierent
elementary particles correspond to dierent vibration modes of the string, in much the same
way as dierent minimal notes correspond to dierent vibrational modes of musical string
instruments. The ve dierent string theories are dierent aspects of a more fundamental
theory, called M-theory [45]. In the latest developments, one deals with branes, which are
classical solutions of the equations of motion of the low-energy string eective action, that
correspond to new non-perturbative states of string theory, break half of the supersymme-
try, and are required by T-duality in theories with open strings. They have the peculiar
property that open strings have their end-points attached to them. With the language of
pseudo-Riemannian geometry, branes are timelike surfaces embedded into bulk spacetime
[32]. According to this picture, gravity lives on the bulk, while standard-model gauge elds
are conned on the brane. For branes, the normal vector N is spacelike with respect to the
bulk metric G
AB
, i.e.,
G
AB
N
A
N
B
= N
C
N
C
> 0. (186)
54
The action functional S splits into the sum [32] (g

(x) being the brane metric)

S = S
4
[g

(x)] + S
5
[G
AB
(X)], (187)
while the eective action [55] is formally given by
e
i
=
_
DG
AB
(X) e
iS
g.f. term. (188)
In the functional integral, the gauge-xed action reads as (here there is summation as well
as integration over repeated indices)
S
g.f.
= S
4
+ S
5
+
1
2
F
A

AB
F
B
+
1
2

, (189)
where F
A
and

are bulk and brane gauge-xing functionals, respectively, while

AB
and

are non-singular matrices of gauge parameters. The gauge-invariance properties of

bulk and brane action functionals can be expressed by saying that there exist vector elds
on the space of histories such that (cf. Eq. (180))
R
B
S
5
= 0, R

S
4
= 0, (190)
whose Lie brackets obey a relation formally analogous to Eq. (181) for ordinary type-I
theories, i.e.,
[R
B
, R
D
] = C
A
BD
R
A
, [R

, R

] = C

. (191)
The bulk and brane ghost operators are therefore
Q
A
B
= R
B
F
A
= F
A
,a
R
a
B
, (192)
J

= R

,i
R
i

, (193)
respectively. The full bulk integration means integrating rst with respect to all bulk metrics
G
AB
inducing on the boundary M the given brane metric g

(x), and then integrating with

respect to all brane metrics. Thus, one rst evaluates the cosmological wave function of the
bulk spacetime [32], i.e.,

Bulk
=
_
G
AB
[M]=g

(G
AB
, S
C
, T
D
)e
i

S
5
, (194)
where is taken to be a suitable measure, the S
C
, T
D
are ghost elds, and (of course, S
A
here diers from the symbol for the action in eq. (103))

S
5
S
5
[G
AB
] +
1
2
F
A

AB
F
B
+ S
A
Q
A
B
T
B
. (195)
55
Eventually, the eective action results from
e
i
=
_
(g

)e
i

S
4

Bulk
, (196)
where is another putative measure,

and

are brane ghost elds, and

S
4
S
4
+
1
2

. (197)
XX. BULK AND BRANE BRST TRANSFORMATIONS
This scheme is invariant under innitesimal BRST transformations on the bulk, given by
G
a
= R
a
A
T
A
, (198)
S
A
=
AB
F
B
, (199)
T
A
=
1
2
C
A
BD
T
B
T
D
, (200)
where T
A
= T
A
, T
A
T
B
= T
B
T
A
, as well as under formally analogous BRST
transformations on the brane, i.e.
g
i
= R
i

, (201)

, (202)

=
1
2
C

, (203)
where

.
XXI. NEW PERSPECTIVES IN THE SPECTRAL ASYMPTOTICS OF EU-
CLIDEAN QUANTUM GRAVITY
Since the early eighties there has been a substantial revival of interest in quantum cos-
mology, motivated by the hope of obtaining a complete picture of how the universe could
arise and evolve [46, 47, 48, 49]. By complete we here mean a theoretical description where,
by virtue of the guiding principles of physics and mathematics, both the dierential equa-
tions of the theory and the associated boundary (and initial) conditions are fully specied.
Even though modern theoretical cosmology deals with yet other deep issues such as dark
56
matter, dark energy [50, 51] and cosmic strings [52], the eort of formulating the appropriate
boundary conditions for the quantum state of the universe [53], or at least for its (one-loop)
semiclassical approximation, plays again a key role, since the universe might have had a
semiclassical origin [54], and the various orders in h in the loop expansion describe the
departure from the underlying classical dynamics.
The physical motivations of our work result therefore from the following active areas of
research:
(i) Functional integrals and space-time approach to quantum eld theory [55].
(ii) Attempt to derive the whole set of physical laws from invariance principles [56].
(iii) How to derive the early universe evolution from quantum physics; how to make sense
of a wave function of the universe and of HartleHawking quantum cosmology [53, 57].
(iv) Spectral theory and its physical applications, including functional determinants in one-
loop quantum theory and hence the rst corrections to classical dynamics [58].
The boundary conditions that we study are part of a unied scheme for Maxwell, Yang
Mills and Quantized General Relativity at one loop, i.e. [59]
_
/
_
B
= 0, (204)
_
(A)
_
B
= 0, (205)
[]
B
= 0. (206)
With our notation, is a projector acting on the gauge eld /, is the gauge-xing func-
tional, is the full set of ghost elds [60]. Both equation (204) and (205) are preserved under
innitesimal gauge transformations provided that the ghost obeys homogeneous Dirichlet
conditions as in (206). For gravity, we choose so as to have an operator P of Laplace type
on metric perturbations in the one-loop Euclidean theory.
XXII. EIGENVALUE CONDITIONS FOR SCALAR MODES
On the Euclidean 4-ball, we expand metric perturbations h

in terms of scalar, transverse

vector, transverse-traceless tensor harmonics on S
3
. For vector, tensor and ghost modes,
57
boundary conditions reduce to Dirichlet or Robin [61]. For scalar modes, one nds eventually
the eigenvalues E = x
2
from the roots x of [61]
J

n
(x)
n
x
J
n
(x) = 0, (207)
J

n
(x) +
_

x
2

n
x
_
J
n
(x) = 0. (208)
Note that both x and x solve the same equation. For example, at small n and large x, the
roots of Eq. (208) with + sign in front of
n
x
read as (here s = 0, 1, ..., )
x(s, n) (s, n)
_
1 +

1

2
(s, n)
+

2

4
(s, n)
+

3

6
(s, n)
+ O(
8
)
_
, (209)
where
(s, n)
_
s +
n
2
+
3
4
_
, (210)
and (having dened m 4n
2
)

1
(m)
(m17)
8
, (211)

2
(m)
3455
384
+ 2m
1/2
+
67
192
m
7
384
m
2
, (212)

3
(m) =
1117523
15360

115
4
m
1/2

5907
5120
m+
3
4
m
3/2
+
421
3072
m
2

83
15360
m
3
, (213)
as has been found in [62].
XXIII. FOUR GENERALIZED -FUNCTIONS FOR SCALAR MODES
From Eqs. (207) and (208) we obtain the following integral representations of the resulting
-functions upon exploiting the Cauchy theorem and rotation of contour [61, 62]:

A,B
(s)
(sin s)

n=3
n
(2s2)
_

0
dz z
2s

z
log F

A,B
(zn), (214)
where (here
+
n,

n + 2)
F

A
(zn) z

_
znI

n
(zn) nI
n
(zn)
_
, (215)
F

B
(zn) z

_
znI

n
(zn) +
_
(zn)
2
2
n
_
I
n
(zn)
_
, (216)
I
n
being the modied Bessel functions of rst kind. Regularity at the origin is easily proved
in the elliptic sectors, corresponding to

A
(s) and

B
(s).
58
XXIV. REGULARITY OF
+
B
AT s = 0
We now dene (1 + z
2
)
1/2
and consider the uniform asymptotic expansion (away
from = 1, with notation as in [61, 62])
z

+
F
+
B
(zn)
e
n()
h(n)

(1
2
)

_
_
1 +

j=1
r
j,+
()
n
j
_
_
, (217)
the functions r
j,+
being obtained from the Olver polynomials for the uniform asymptotic
expansion of I
n
and I

n
[63]. On splitting
_
1
0
d =
_

0
d +
_
1

d with small, we get an

asymptotic expansion of the l.h.s. by writing, in the rst interval on the r.h.s.,
log
_
_
1 +

j=1
r
j,+
()
n
j
_
_

j=1
R
j,+
()
n
j
, (218)
and then computing
C
j
()
R
j,+

= (1 )
j1
4j

a=j1
K
(j)
a

a
. (219)
The integral
_
1

d is instead found to yield a vanishing contribution in the 1 limit [62].

Remarkably, by virtue of the spectral identity
g(j)
4j

a=j
(a + 1)
(a j + 1)
K
(j)
a
= 0, (220)
which holds j = 1, ..., , we nd
lim
s0
s
+
B
(s) =
1
6
12

a=3
a(a 1)(a 2)K
(3)
a
= 0, (221)
and

+
B
(0) =
5
4
+
1079
240

1
2
12

a=2
(a)K
(3)
a
+

j=1
f(j)g(j) =
296
45
, (222)
where
(a)
1
6
(a + 1)
(a 2)
_
log(2)
(6a
2
9a + 1)
4
(a 2)
(a + 1)
+ 2(a + 1) (a 2) (4)
_
, (223)
f(j)
(1)
j
j!
_
1 2
2j
+
R
(j 2)(1
j,3
) +
j,3
_
. (224)
The spectral cancellation (220) achieves three goals: (i) Vanishing of log 2 coecient in Eq.
(222); (ii) Vanishing of

j=1
f(j)g(j) in Eq. (222); (iii) Regularity at the origin of
+
B
.
59
To cross-check our analysis, we evaluate r
j,+
() r
j,
() and hence obtain R
j,+
()
R
j,
() for all j. Only j = 3 contributes to

B
(0), and we nd

+
B
(0) =

B
(0)
1
24
4

l=1
(l + 1)
(l 2)
_
(l + 2)
1
(l + 1)
_

(3)
2l+1
=
206
45
+ 2 =
296
45
, (225)
in agreement with Eq. (222), where
(3)
2l+1
are the four coecients on the right-hand side of

(R
3,+
R
3,
) = (1
2
)
4
_
80
3
24
5
+ 32
7
8
9
_
. (226)
Within this framework, the spectral cancellation reads as
4

l=1
(l + 1)
(l 2)

(3)
2l+1
= 0, (227)
which is a particular case of
a=amax(j)

a=a
min
(j)
((a + 1)/2)
((a + 1)/2 j)

(j)
a
= 0. (228)
Interestingly, the full (0) value for pure gravity (i.e. including the contribution of tensor,
vector, scalar and ghost modes) is then found to be positive: (0) =
142
45
[62], which suggests
a quantum avoidance of the cosmological singularity driven by full dieomorphism invariance
of the boundary-value problem for one-loop quantum theory [62].
XXV. SELECTED OPEN PROBLEMS
Several open problems should be brought to the attention of the reader, and are as follows.
(i) We have encountered in sections 21-24 a boundary-value problem where the generalized
-function remains well dened, even though the Mellin transform relating -function to
heat kernel does not exist (see further comments below), since strong ellipticity is violated
[59] (see also [64]). Are the spectral cancellations (220) and (228) a peculiar property of
the Euclidean 4-ball, or can they be extended to more general Riemannian manifolds with
non-empty boundary?
(ii) What is the deeper underlying reason for nding
+
B
(0)

B
(0) = 2? Is it possible to
foresee a geometrical or topological or group-theoretical origin of this result?
60
(iii) Is it correct to say that our positive (0) value for pure gravity engenders a quantum
avoidance of the cosmological singularity at one-loop level? [54, 62] Does the result remain
true in higher-loop calculations or on using other regularization techniques for the one-loop
correction?
(iv) The whole scheme might be relevant for AdS/CFT in light of a profound link between
AdS/CFT and the HartleHawking wave function of the universe [65].
(v) What happens if one considers instead non-local boundary data, e.g. those giving rise
to surface states for the Laplacian? [56, 66, 67]
As far as item (i) is concerned, we should add what follows. The integral representation
(214) of the generalized -function is legitimate because, for any xed n, there is a count-
able innity of roots x
j
and x
j
of eqs. (207) and (208), and they grow approximately
linearly with the integer j counting such roots. The functions F

A
and F

B
admit therefore
a canonical-product representation [68] which ensures that the integral representation (214)
reproduces the standard denition of generalized -function [61]. Furthermore, even though
the Mellin transform relating -function to integrated heat kernel cannot be exploited when
strong ellipticity is not fullled, it remains possible to dene a generalized -function. For
this purpose, a weaker assumption provides a sucient condition, i.e. the existence of a
sector in the complex plane free of eigenvalues of the leading symbol of the dierential op-
erator under consideration [61, 62]. To make sure we have not overlooked some properties
of the spectrum, we have been looking for negative eigenvalues or zero-modes, but nding
none. Indeed, negative eigenvalues E would imply purely imaginary roots x = iy of eq.
(208), but such roots do not exist, as one can check both numerically and analytically;
zero-modes would be non-trivial eigenfunctions belonging to zero-eigenvalues, but all modes
(tensor, vector, scalar and ghost modes) are combinations of regular Bessel functions [61]
(since we require regularity at the origin of the left-hand side of eq. (204)) for which this
is impossible. As far as we can see, we still nd sources of singularities at the origin in the
generalized -function resulting from lack of strong ellipticity, but the particular symmetries
of the Euclidean 4-ball background reduce them to the four terms in Eq. (227), which add
up to zero despite two of them are non-vanishing.
In [62] we have proposed to interpret the result (0) =
142
45
for pure gravity as an indication
61
that full dieomorphism invariance of the boundary-value problem engenders a
quantum avoidance of the cosmological singularity. Indeed, on the one hand, the
work by Schleich [69] had found that, on restricting the functional integral to transverse-
traceless perturbations, the one-loop semiclassical approximation to the wave function of the
universe diverges at small volumes, at least for the boundary geometry of a three-sphere.
The divergence of the wave functional does not imply, by itself, that the probability density
of the wave functional diverges at small volumes, since the probability density p[h] on the
space of wave functionals [h] is given by p[h] = m[h][[
2
[h], where m[h] is the measure
on this space, the scaling of which is not known in general. On the other hand, in our
manifestly covariant evaluation of the one-loop functional integral for the wave function of
the universe, it seems incorrect to assume that the measure m[h] scales in such a way as to
cancel exactly the contribution of the squared modulus of , which is proportional to the
three-sphere radius raised to the power 2(0). Thus, we nd that our one-loop wave function
of the universe vanishes at small volume. The normalizability condition of the wave function
in the limit of small three-geometry, which is weaker than requiring it should vanish in this
limit, was instead formulated and studied in [70].
The years to come will hopefully tell us whether our calculations may be viewed as a rst
step towards nding under which conditions a quantum theory of gravity is singularity free
in cosmology [71]. For this purpose, it might also be interesting to study dieomorphism-
invariant boundary conditions for f(R) theories of gravity, recently studied at one-loop level
on manifolds without boundary [72].
On the non-perturbative side, encouraging progress has been made towards nding cosmo-
logical applications of non-perturbative quantum gravity via renormalization-group methods
[27], including, in particular, theoretical models that might account for the accelerated ex-
pansion of the universe [73] and for at rotation curves of galaxies [74].
Acknowledgments
G. Esposito is grateful to the Dipartimento di Scienze Fisiche of Federico II University,
Naples, for hospitality and support, and to Professor C. DeWittMorette for warm approval
of the present project. His work has been partially supported by PRIN SINTESI. Springer
has kindly granted permission to reprint, from section 2 to 17, the 1978 Carg`ese Lectures by
62
Professor B.S. DeWitt. Along the years, G. Esposito has learned a lot from many colleagues,
e.g. I.G. Avramidi, P.D. DEath, A.Yu. Kamenshchik, K. Kirsten, and the interaction with
many Ph.D. students has been very helpful in the course of preparing this teaching-oriented
review.
APPENDIX A: LIE GROUPS
A Lie group is a group G which is also a manifold with a C

structure such that the

maps
(x, y) xy
x x
1
are C

functions. It is indeed enough to assume that (x, y) xy

1
, or (x, y) xy, are
C

. Relevant examples of Lie groups are as follows [75].

(1) The space R
n
endowed with the addition +.
(2) The circle S
1
dened as the quotient space R/Z.
(3) If G and H are Lie groups, their product GH is also a Lie group.
(4) The torus S
1
S
1
.
(5) The general linear group GL(n, R), i.e. the group of all nn nonsingular real matrices.
This is a subset of R
n
2
.
(6) The orthogonal group
O(n)
_
A GL(n, R) : AA
t
= I
_
. (A1)
Here, with respect to the usual base of R
n
, the matrix A represents a linear map which is
an isometry, i.e. it preserves the norm and the inner product.
(7) If H G is a subgroup of G and also a submanifold of G, then H is a Lie group. Thus,
for example, S
1
R
2
is a Lie group because
S
1
=
_
z C : zz = x
2
+ y
2
= 1
_
. (A2)
63
(8) The three-sphere S
3
is the Lie group of unit norm quaternions (among all the S
n
, only
S
1
and S
3
admit a Lie-group structure). On introducing the three symbols i, j, k such that
i
2
= j
2
= k
2
= 1, (A3)
ij = ji = k, jk = kj = i, ki = ik = j, (A4)
a quaternion can be expressed in the form
x = x
1
+ x
2
i + x
3
j + x
4
k, (A5)
with (x
1
, x
2
, x
3
, x
4
) R
4
, so that the complex conjugate quaternion reads as
x = x
1
x
2
i x
3
j x
4
k, (A6)
and the equation dening S
3
can be indeed satised, i.e.
xx =
4

i=1
x
2
i
= 1. (A7)
(9) The group SO(n) of all orthogonal matrices with unit determinant, i.e.
SO(n) A O(n) : detA = 1 , (A8)
is a Lie group.
(10) The group E(n) of all isometries of R
n
is a Lie group. Every element of E(n) can be
written uniquely as A where A O(n), and is a translation
(x) =
a
(x) = x + a. (A9)
Note that E(n) ,= O(n) R
n
because translations and orthogonal transformations do not
commute. It is however true that O(n) is dieomorphic to O(n) R
n
.
Given a Lie group G, its Lie algebra g is the tangent space to G at its identity element:
g = T
e
G. In general, a Lie algebra is a nite-dimensional vector space V , endowed with an
antisymmetric bilinear map [ , ]
[X, Y ] = [Y, X] X, Y V (A10)
which satises the Jacobi identity
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0 X, Y, Z V. (A11)
64
An important theorem asserts that, in the nite-dimensional case, a Lie algebra is always
isomorphic to the Lie algebra g of a nite-dimensional Lie group G. With a standard
notation, one writes gl(n, R) for the Lie algebra of GL(n, R) and o(n) for the Lie algebra of
O(n). Every A g is dened by its value at the unit element of G. Given a basis e
i
in g,
one has the Lie-bracket relations (see eq. (14))
[e
i
, e
j
] = f
k
ij
e
k
, (A12)
where f
k
ij
are the structure constants of G.
In quantum gravity and quantum Yang-Mills theories one deals however with innite-
dimensional Lie groups (also called pseudo-groups in the literature). The adjoint represen-
tation of the dieomorphism group is provided by a contravariant vector eld X

, as can be
seen from the transformation law [76]
X

=
_
d
n
x

_
d
n
x

= X

+ X

,
. (A13)
The coadjoint representation is instead provided by a covariant vector density of unit weight
according to [76]
Y

=
_
d
n
x

_
d
n
x

= (Y

)
,
Y

,
. (A14)
We refer the reader to the work in [77] for recent results on gravitation as gauge theory of
the dieomorphism group, while deformations of dieomorphisms are studied in detail in
[78, 79].
[1] S.W. Hawking and R. Penrose, Proc. R. Soc. Lond. A314, 529 (1970).
[2] J.R. Klauder, gr-qc/0612168.
[3] R. Geroch, Ann. Phys. (N.Y.) 62, 582 (1971).
[4] A. Ashtekar, Phys. Rev. Lett. 46, 573 (1981).
[5] B.S. DeWitt, Phys. Rev. 160, 1113 (1967).
[6] C.J. Isham and A.C. Kakas, Class. Quantum Grav. 1, 621 (1984).
[7] C.J. Isham and A.C. Kakas, Class. Quantum Grav. 1, 633 (1984).
[8] A. Ashtekar, Phys. Rev. Lett. 57, 2244 (1986).
[9] G. Esposito, G. Gionti and C. Stornaiolo, Nuovo Cim. B110, 1137 (1995).
65
[10] G.E. Volovik, The Universe in a Helium Droplet, Int. Ser. Monogr. Phys. 117, 1 (2006).
[11] C.W. Misner, Rev. Mod. Phys. 29, 497 (1957).
[12] B.S. DeWitt, Phys. Rev. 162, 1195 (1967).
[13] G. t Hooft and M. Veltman, Ann. Inst. H. Poincare A20, 69 (1974).
[14] M.H. Goro and A. Sagnotti, Nucl. Phys. B266, 709 (1986).
[15] I.G. Avramidi, Nucl. Phys. B355, 712 (1991).
[16] G.A. Vilkovisky, Class. Quantum Grav. 9, 895 (1992).
[17] F. Canfora, Nucl. Phys. B731, 389 (2005).
[18] G.W. Gibbons and S.W. Hawking, Phys. Rev. D15, 2752 (1977).
[19] S.W. Hawking, Phys. Rev. D18, 1747 (1978).
[20] P. Menotti and A. Pelissetto, Phys. Rev. D35, 1194 (1987).
[21] M. Carfora and A. Marzuoli, Class. Quantum Grav. 9, 595 (1992).
[22] C. Rovelli and L. Smolin, Phys. Rev. Lett. 61, 1155 (1988).
[23] C. Rovelli and L. Smolin, Nucl. Phys. B331, 80 (1990).
[24] J. Gracia-Bondia, hep-th/0206006.
[25] C.J. Isham, Class. Quantum Grav. 6, 1509 (1989).
[26] J.A. Wheeler, Ann. Phys. (N.Y.) 2, 604 (1957).
[27] M. Reuter, Phys. Rev. D57, 971 (1998).
[28] O. Lauscher and M. Reuter, hep-th/0511260.
[29] K. Stelle, Phys. Rev. D16, 953 (1977).
[30] E. Witten, hep-th/0212247.
[31] G.T. Horowitz, New J. Phys. 7, 201 (2005).
[32] A.O. Barvinsky, Phys. Rev. D74, 084033 (2006).
[33] D.Z. Freedman, P. van Nieuwenhuizen and S. Ferrara, Phys. Rev. D13, 3214 (1976).
[34] P. van Nieuwenhuizen, Phys. Rep. 68, 189 (1981).
[35] G. Gionti, gr-qc/9812080.
[36] J. Ambjorn, M. Carfora, D. Gabrielli, A. Marzuoli, Nucl. Phys. B542, 349 (1999).
[37] R. Loll, arXiv:0711.0273.
[38] A. Kheyfets, N.J. Lafave, W.A. Miller, Class. Quantum Grav. 6, 659 (1989).
[39] R. Penrose and M.A.H. MacCallum, Phys. Rep. 6, 241 (1972).
[40] R. Penrose, Class. Quantum Grav. 16, A113 (1999).
66
[41] B.S. DeWitt, Dynamical Theory of Groups and Fields (Gordon & Breach, New York, 1965).
[42] L. Lorenz, Phil. Mag. 34, 287 (1867).
[43] W. Rarita and J. Schwinger, Phys. Rev. 60, 61 (1941).
[44] S. Deser and B. Zumino, Phys. Lett. B62, 335 (1976).
[45] K. Becker, M. Becker and J.H. Schwarz, String Theory and M-Theory: a Modern Introduction
(Cambridge University Press, Cambridge, 2007).
[46] S.W. Hawking, Pont. Acad. Sci. Scr. Varia 48, 563 (1982).
[47] A. Vilenkin, Phys. Rev. D30, 509 (1984).
[48] A. Vilenkin, Phys. Rev. D33, 3560 (1986).
[49] A. Vilenkin, Phys. Rev. D37, 888 (1988).
[50] A.Yu. Kamenshchik, U. Moschella and V. Pasquier, Phys. Lett. B511, 265 (2001).
[51] S. Capozziello, S. Nojiri and S.D. Odintsov, Phys. Lett. B632, 597 (2006).
[52] A. Vilenkin, hep-th/0508135.
[53] S.W. Hawking, Nucl. Phys. B239, 257 (1984).
[54] S.W. Hawking, Phys. Scripta T117, 49 (2005).
[55] B.S. DeWitt, The Global Approach to Quantum Field Theory, Int. Ser. Monogr. Phys. 114, 1
(2003).
[56] G. Esposito, Int. J. Mod. Phys. A15, 4539 (2000).
[57] J.B. Hartle and S.W. Hawking, Phys. Rev. D28, 2960 (1983).
[58] G. Esposito, Quantum Gravity, Quantum Cosmology and Lorentzian Geometries, Lecture
Notes in Physics, New Series m, Vol. m12 (Springer, Berlin, 1994).
[59] I.G. Avramidi and G. Esposito, Commun. Math. Phys. 200, 495 (1999).
[60] G. Esposito, A.Yu. Kamenshchik and G. Pollifrone, Euclidean Quantum Gravity on Manifolds
with Boundary, Vol. 85 of Fundamental Theories of Physics (Kluwer, Dordrecht, 1997).
[61] G. Esposito, G. Fucci, A.Yu. Kamenshchik and K. Kirsten, Class. Quantum Grav. 22, 957
(2005).
[62] G. Esposito, G. Fucci, A.Yu. Kamenshchik and K. Kirsten, JHEP 0509:063 (2005).
[63] F.W.J. Olver, Phil. Trans. R. Soc. Lond. A247, 328 (1954).
[64] J.S. Dowker and K. Kirsten, Class. Quantum Grav. 14, L169 (1997).
[65] G.T. Horowitz and J. Maldacena, JHEP 0402:008 (2004).
[66] M. Schr oder, Rep. Math. Phys. 27, 259 (1989).
67
[67] G. Esposito, Class. Quantum Grav. 16, 1113 (1999).
[68] L.V. Ahlfors, Complex Analysis (McGraw-Hill, New York, 1966).
[69] K. Schleich, Phys. Rev. D32, 1889 (1985).
[70] A.O. Barvinsky and A.Yu. Kamenshchik, Class. Quantum Grav. 7, L181 (1990).
[71] C. Kiefer, Ann. der Phys. 15, 129 (2005).
[72] G. Cognola, E. Elizalde, S. Nojiri, S.D. Odintsov and S. Zerbini, JCAP 02:010 (2005).
[73] A. Bonanno, G. Esposito, C. Rubano and P. Scudellaro, Class. Quantum Grav. 23, 3103
(2006).
[74] G. Esposito, C. Rubano and P. Scudellaro, arXiv:0709.1403.
[75] M. Spivak, A Comprehensive Introduction to Dierential Geometry, Vol. 1 (Publish or Perish,
Houston, 1970).
[76] B.S. DeWitt, in Relativity, Groups and Topology II, eds. B.S. DeWitt and R. Stora (North-
Holland, Amsterdam, 1984).
[77] Y.M. Cho, K.S. Soh, J.H. Yoon, Q-Han Park, Phys. Lett. B286, 251 (1992).
[78] P. Aschieri, C. Blohmann, M. Dimitrijevic, F. Meyer, P. Schupp and J. Wess, Class. Quantum
Grav. 22, 3511 (2005).
[79] P. Aschieri, M. Dimitrijevic, F. Meyer and J. Wess, Class. Quantum Grav. 23, 1883 (2006).
68